nn.py 379.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
Y
Yu Yang 已提交
14
"""
15
All layers just related to the neural network.
Y
Yu Yang 已提交
16 17
"""

18 19
from __future__ import print_function

20
import numpy as np
S
sneaxiy 已提交
21
import six
P
peizhilin 已提交
22
import os
S
sneaxiy 已提交
23
import inspect
Y
Yu Yang 已提交
24
from ..layer_helper import LayerHelper
25
from ..initializer import Normal, Constant, NumpyArrayInitializer
S
sneaxiy 已提交
26
from ..framework import Variable, OpProtoHolder
Y
yangyaming 已提交
27
from ..param_attr import ParamAttr
S
sneaxiy 已提交
28
from .layer_function_generator import autodoc, templatedoc, _generate_doc_string_
29
from .tensor import concat, assign
30
from . import utils
F
fengjiayi 已提交
31
from .. import unique_name
32
from functools import reduce
33
from .. import core
X
Xin Pan 已提交
34
from ..imperative import layers
Y
Yu Yang 已提交
35 36

__all__ = [
X
Xin Pan 已提交
37 38 39 40 41 42 43 44 45 46
    'fc',
    'embedding',
    'dynamic_lstm',
    'dynamic_lstmp',
    'dynamic_gru',
    'gru_unit',
    'linear_chain_crf',
    'crf_decoding',
    'cos_sim',
    'cross_entropy',
47
    'bpr_loss',
X
Xin Pan 已提交
48 49 50 51 52 53 54 55 56 57
    'square_error_cost',
    'chunk_eval',
    'sequence_conv',
    'conv2d',
    'conv3d',
    'sequence_pool',
    'sequence_softmax',
    'softmax',
    'pool2d',
    'pool3d',
58 59
    'adaptive_pool2d',
    'adaptive_pool3d',
X
Xin Pan 已提交
60
    'batch_norm',
H
heqiaozhi 已提交
61
    'data_norm',
X
Xin Pan 已提交
62 63 64 65 66 67
    'beam_search_decode',
    'conv2d_transpose',
    'conv3d_transpose',
    'sequence_expand',
    'sequence_expand_as',
    'sequence_pad',
Y
Yibing Liu 已提交
68
    'sequence_unpad',
X
Xin Pan 已提交
69 70 71 72 73 74 75 76
    'lstm_unit',
    'reduce_sum',
    'reduce_mean',
    'reduce_max',
    'reduce_min',
    'reduce_prod',
    'sequence_first_step',
    'sequence_last_step',
Y
Yibing Liu 已提交
77
    'sequence_slice',
X
Xin Pan 已提交
78 79 80 81 82 83 84 85 86 87 88 89
    'dropout',
    'split',
    'ctc_greedy_decoder',
    'edit_distance',
    'l2_normalize',
    'matmul',
    'topk',
    'warpctc',
    'sequence_reshape',
    'transpose',
    'im2sequence',
    'nce',
90
    'sampled_softmax_with_cross_entropy',
X
Xin Pan 已提交
91 92 93 94 95
    'hsigmoid',
    'beam_search',
    'row_conv',
    'multiplex',
    'layer_norm',
D
Dun 已提交
96
    'group_norm',
X
Xin Pan 已提交
97 98 99 100 101 102 103 104 105 106 107 108 109
    'softmax_with_cross_entropy',
    'smooth_l1',
    'one_hot',
    'autoincreased_step_counter',
    'reshape',
    'squeeze',
    'unsqueeze',
    'lod_reset',
    'lrn',
    'pad',
    'pad_constant_like',
    'label_smooth',
    'roi_pool',
J
jerrywgz 已提交
110
    'roi_align',
X
Xin Pan 已提交
111 112 113 114
    'dice_loss',
    'image_resize',
    'image_resize_short',
    'resize_bilinear',
115
    'resize_nearest',
X
Xin Pan 已提交
116 117 118 119 120 121
    'gather',
    'scatter',
    'sequence_scatter',
    'random_crop',
    'mean_iou',
    'relu',
C
chengduo 已提交
122
    'selu',
X
Xin Pan 已提交
123 124 125
    'log',
    'crop',
    'rank_loss',
M
minqiyang 已提交
126
    'margin_rank_loss',
X
Xin Pan 已提交
127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169
    'elu',
    'relu6',
    'pow',
    'stanh',
    'hard_sigmoid',
    'swish',
    'prelu',
    'brelu',
    'leaky_relu',
    'soft_relu',
    'flatten',
    'sequence_mask',
    'stack',
    'pad2d',
    'unstack',
    'sequence_enumerate',
    'expand',
    'sequence_concat',
    'scale',
    'elementwise_add',
    'elementwise_div',
    'elementwise_sub',
    'elementwise_mul',
    'elementwise_max',
    'elementwise_min',
    'elementwise_pow',
    'uniform_random_batch_size_like',
    'gaussian_random',
    'sampling_id',
    'gaussian_random_batch_size_like',
    'sum',
    'slice',
    'shape',
    'logical_and',
    'logical_or',
    'logical_xor',
    'logical_not',
    'clip',
    'clip_by_norm',
    'mean',
    'mul',
    'sigmoid_cross_entropy_with_logits',
    'maxout',
J
JiabinYang 已提交
170
    'space_to_depth',
W
whs 已提交
171
    'affine_grid',
S
sneaxiy 已提交
172
    'sequence_reverse',
173
    'affine_channel',
B
barrierye 已提交
174
    'similarity_focus',
M
minqiyang 已提交
175
    'hash',
D
dengkaipeng 已提交
176
    'grid_sampler',
G
gmcather 已提交
177 178
    'log_loss',
    'add_position_encoding',
Q
Qiao Longfei 已提交
179
    'bilinear_tensor_product',
C
chengduo 已提交
180 181
    'merge_selected_rows',
    'get_tensor_from_selected_rows',
P
phlrain 已提交
182
    'lstm',
S
shippingwang 已提交
183
    'shuffle_channel',
S
sneaxiy 已提交
184
    'py_func',
185
    'psroi_pool',
H
heqiaozhi 已提交
186
    'teacher_student_sigmoid_loss',
M
minqiyang 已提交
187
    'huber_loss',
Z
zhaozhehao 已提交
188
    'tree_conv',
C
ceci3 已提交
189
    'npair_loss',
Y
Yu Yang 已提交
190 191
]

J
jerrywgz 已提交
192 193
kIgnoreIndex = -100

Y
Yu Yang 已提交
194 195 196 197 198 199 200

def fc(input,
       size,
       num_flatten_dims=1,
       param_attr=None,
       bias_attr=None,
       act=None,
J
Jacek Czaja 已提交
201
       is_test=False,
202
       name=None):
Y
Yu Yang 已提交
203
    """
204
    **Fully Connected Layer**
Y
Yu Yang 已提交
205

206 207 208 209 210 211 212 213
    This function creates a fully connected layer in the network. It can take
    multiple tensors as its inputs. It creates a variable called weights for
    each input tensor, which represents a fully connected weight matrix from
    each input unit to each output unit. The fully connected layer multiplies
    each input tensor with its coresponding weight to produce an output Tensor.
    If multiple input tensors are given, the results of multiple multiplications
    will be sumed up. If bias_attr is not None, a bias variable will be created
    and added to the output. Finally, if activation is not None, it will be applied
F
fengjiayi 已提交
214
    to the output as well.
C
caoying03 已提交
215

C
caoying03 已提交
216
    This process can be formulated as follows:
217 218 219

    .. math::

220
        Out = Act({\sum_{i=0}^{N-1}X_iW_i + b})
221 222 223

    In the above equation:

C
caoying03 已提交
224 225 226 227
    * :math:`N`: Number of the input.
    * :math:`X_i`: The input tensor.
    * :math:`W`: The weights created by this layer.
    * :math:`b`: The bias parameter created by this layer (if needed).
228
    * :math:`Act`: The activation function.
C
caoying03 已提交
229
    * :math:`Out`: The output tensor.
Y
Yu Yang 已提交
230 231

    Args:
R
ranqiu 已提交
232 233 234 235 236 237 238 239 240 241
        input (Variable|list of Variable): The input tensor(s) of this layer, and the dimension of
            the input tensor(s) is at least 2.
        size(int): The number of output units in this layer.
        num_flatten_dims (int, default 1): The fc layer can accept an input tensor with more than
            two dimensions. If this happens, the multidimensional tensor will first be flattened
            into a 2-dimensional matrix. The parameter `num_flatten_dims` determines how the input
            tensor is flattened: the first `num_flatten_dims` (inclusive, index starts from 1)
            dimensions will be flatten to form the first dimension of the final matrix (height of
            the matrix), and the rest `rank(X) - num_flatten_dims` dimensions are flattened to
            form the second dimension of the final matrix (width of the matrix). For example, suppose
H
haowang101779990 已提交
242
            `X` is a 5-dimensional tensor with a shape [2, 3, 4, 5, 6], and `num_flatten_dims` = 3.
R
ranqiu 已提交
243 244 245 246
            Then, the flattened matrix will have a shape [2 x 3 x 4, 5 x 6] = [24, 30].
        param_attr (ParamAttr|list of ParamAttr, default None): The parameter attribute for learnable
            parameters/weights of this layer.
        bias_attr (ParamAttr|list of ParamAttr, default None): The parameter attribute for the bias
247 248
            of this layer. If it is set to False, no bias will be added to the output units.
            If it is set to None, the bias is initialized zero. Default: None.
R
ranqiu 已提交
249
        act (str, default None): Activation to be applied to the output of this layer.
J
Jacek Czaja 已提交
250
        is_test(bool): A flag indicating whether execution is in test phase.
R
ranqiu 已提交
251
        name (str, default None): The name of this layer.
Y
Yu Yang 已提交
252

253
    Returns:
F
fengjiayi 已提交
254
        Variable: The transformation result.
255 256

    Raises:
C
caoying03 已提交
257
        ValueError: If rank of the input tensor is less than 2.
258 259 260 261

    Examples:
        .. code-block:: python

F
fengjiayi 已提交
262
          data = fluid.layers.data(name="data", shape=[32, 32], dtype="float32")
263
          fc = fluid.layers.fc(input=data, size=1000, act="tanh")
Y
Yu Yang 已提交
264
    """
C
caoying03 已提交
265

C
caoying03 已提交
266
    helper = LayerHelper("fc", **locals())
Y
Yu Yang 已提交
267 268 269 270

    dtype = helper.input_dtype()

    mul_results = []
271 272
    for input_var, param_attr in helper.iter_inputs_and_params():
        input_shape = input_var.shape
Y
Yu Yang 已提交
273 274 275
        param_shape = [
            reduce(lambda a, b: a * b, input_shape[num_flatten_dims:], 1)
        ] + [size]
Y
ying 已提交
276

Y
Yu Yang 已提交
277
        w = helper.create_parameter(
278
            attr=param_attr, shape=param_shape, dtype=dtype, is_bias=False)
X
Xin Pan 已提交
279
        tmp = helper.create_variable_for_type_inference(dtype)
280
        helper.append_op(
281 282 283
            type="mul",
            inputs={"X": input_var,
                    "Y": w},
284
            outputs={"Out": tmp},
M
mozga-intel 已提交
285 286
            attrs={"x_num_col_dims": num_flatten_dims,
                   "y_num_col_dims": 1})
287 288 289 290
        mul_results.append(tmp)

    if len(mul_results) == 1:
        pre_bias = mul_results[0]
291
    else:
X
Xin Pan 已提交
292
        pre_bias = helper.create_variable_for_type_inference(dtype)
293
        helper.append_op(
294 295 296
            type="sum",
            inputs={"X": mul_results},
            outputs={"Out": pre_bias},
X
Xin Pan 已提交
297
            attrs={"use_mkldnn": False})
298 299 300 301
    # add bias
    pre_activation = helper.append_bias_op(pre_bias, dim_start=num_flatten_dims)
    # add activation
    return helper.append_activation(pre_activation)
Y
Yu Yang 已提交
302 303


304 305 306
def embedding(input,
              size,
              is_sparse=False,
307
              is_distributed=False,
308 309 310
              padding_idx=None,
              param_attr=None,
              dtype='float32'):
Y
Yu Yang 已提交
311
    """
312 313
    **Embedding Layer**

314
    This layer is used to lookup embeddings of IDs, provided by :attr:`input`, in
315 316
    a lookup table. The result of this lookup is the embedding of each ID in the
    :attr:`input`.
317 318 319

    All the input variables are passed in as local variables to the LayerHelper
    constructor.
Y
Yu Yang 已提交
320 321

    Args:
322 323 324 325 326
        input(Variable): The tensor variable containing the IDs.
        size(tuple|list): The shape of the look up table parameter. It should
            have two elements which indicate the size of the dictionary of
            embeddings and the size of each embedding vector respectively.
        is_sparse(bool): The flag indicating whether to use sparse update.
327
        is_distributed(bool): Whether to run lookup table from remote parameter server.
328 329
        padding_idx(int|long|None): If :attr:`None`, it makes no effect to lookup.
            Otherwise the given :attr:`padding_idx` indicates padding the output
330
            with zeros whenever lookup encounters it in :attr:`input`. If
331
            :math:`padding_idx < 0`, the :attr:`padding_idx` to use in lookup is
332 333
            :math:`size[0] + dim`.
        param_attr(ParamAttr): Parameters for this layer
334
        dtype(np.dtype|core.VarDesc.VarType|str): The type of data : float32, float_16, int etc
Y
Yu Yang 已提交
335

336 337 338
    Returns:
        Variable: The tensor variable storing the embeddings of the \
                  supplied inputs.
Y
Yu Yang 已提交
339

340 341
    Examples:
        .. code-block:: python
Y
Yu Yang 已提交
342

C
chengduoZH 已提交
343
          dict_size = len(dataset.ids)
344
          data = fluid.layers.data(name='ids', shape=[32, 32], dtype='float32')
C
chengduoZH 已提交
345
          fc = fluid.layers.embedding(input=data, size=[dict_size, 16])
Y
Yu Yang 已提交
346 347 348
    """

    helper = LayerHelper('embedding', **locals())
349
    remote_prefetch = is_sparse and (not is_distributed)
Q
Qiao Longfei 已提交
350 351
    if remote_prefetch:
        assert is_sparse is True and is_distributed is False
Y
Yu Yang 已提交
352 353
    w = helper.create_parameter(
        attr=helper.param_attr, shape=size, dtype=dtype, is_bias=False)
X
Xin Pan 已提交
354
    tmp = helper.create_variable_for_type_inference(dtype)
355 356
    padding_idx = -1 if padding_idx is None else padding_idx if padding_idx >= 0 else (
        size[0] + padding_idx)
Y
Yu Yang 已提交
357 358 359 360 361
    helper.append_op(
        type='lookup_table',
        inputs={'Ids': input,
                'W': w},
        outputs={'Out': tmp},
362 363 364
        attrs={
            'is_sparse': is_sparse,
            'is_distributed': is_distributed,
Q
Qiao Longfei 已提交
365
            'remote_prefetch': remote_prefetch,
366 367
            'padding_idx': padding_idx
        })
Y
Yu Yang 已提交
368 369 370
    return tmp


W
wopeizl 已提交
371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386
@templatedoc(op_type="lstm")
def dynamic_lstm(input,
                 size,
                 h_0=None,
                 c_0=None,
                 param_attr=None,
                 bias_attr=None,
                 use_peepholes=True,
                 is_reverse=False,
                 gate_activation='sigmoid',
                 cell_activation='tanh',
                 candidate_activation='tanh',
                 dtype='float32',
                 name=None):
    """
    ${comment}
Y
Yibing Liu 已提交
387

W
wopeizl 已提交
388 389 390 391 392 393 394 395 396 397 398
    Args:
        input (Variable): ${input_comment}
        size (int): 4 * hidden size.
        h_0(Variable): The initial hidden state is an optional input, default is zero.
                       This is a tensor with shape (N x D), where N is the
                       batch size and D is the hidden size.
        c_0(Variable): The initial cell state is an optional input, default is zero.
                       This is a tensor with shape (N x D), where N is the
                       batch size. `h_0` and `c_0` can be NULL but only at the same time.
        param_attr(ParamAttr|None): The parameter attribute for the learnable
                               hidden-hidden weights.
Y
Yu Yang 已提交
399

W
wopeizl 已提交
400 401 402 403
                               - Weights = {:math:`W_{ch}, W_{ih}, \
                                                W_{fh}, W_{oh}`}
                               - The shape is (D x 4D), where D is the hidden
                                 size.
Y
Yu Yang 已提交
404

W
wopeizl 已提交
405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490
                               If it is set to None or one attribute of ParamAttr,
                               dynamic_lstm will create ParamAttr as param_attr.
                               If the Initializer of the param_attr is not set, the
                               parameter is initialized with Xavier. Default: None.
        bias_attr (ParamAttr|None): The bias attribute for the learnable bias
                              weights, which contains two parts, input-hidden
                              bias weights and peephole connections weights if
                              setting `use_peepholes` to `True`.

                              1. `use_peepholes = False`
                                 - Biases = {:math:`b_c, b_i, b_f, b_o`}.
                                 - The shape is (1 x 4D).
                              2. `use_peepholes = True`
                                 - Biases = { :math:`b_c, b_i, b_f, b_o, W_{ic}, \
                                                 W_{fc}, W_{oc}`}.
                                 - The shape is (1 x 7D).

                              If it is set to None or one attribute of ParamAttr,
                              dynamic_lstm will create ParamAttr as bias_attr.
                              If the Initializer of the bias_attr is not set,
                              the bias is initialized zero. Default: None.
        use_peepholes (bool): ${use_peepholes_comment}
        is_reverse (bool): ${is_reverse_comment}
        gate_activation (str): ${gate_activation_comment}
        cell_activation (str): ${cell_activation_comment}
        candidate_activation (str): ${candidate_activation_comment}
        dtype (str): Data type. Choices = ["float32", "float64"], default "float32".
        name (str|None): A name for this layer(optional). If set None, the layer
                         will be named automatically.

    Returns:
        tuple: The hidden state, and cell state of LSTM. The shape of both \
        is (T x D), and lod is the same with the `input`.

    Examples:
        .. code-block:: python

            hidden_dim = 512
            forward_proj = fluid.layers.fc(input=input_seq, size=hidden_dim * 4,
                                           bias_attr=False)
            forward, _ = fluid.layers.dynamic_lstm(
                input=forward_proj, size=hidden_dim * 4, use_peepholes=False)
    """
    assert bias_attr is not False, "bias_attr should not be False in dynamic_lstmp."
    helper = LayerHelper('lstm', **locals())
    size = size // 4
    weight = helper.create_parameter(
        attr=helper.param_attr, shape=[size, 4 * size], dtype=dtype)
    bias_size = [1, 7 * size]
    if not use_peepholes:
        bias_size[1] = 4 * size
    bias = helper.create_parameter(
        attr=helper.bias_attr, shape=bias_size, dtype=dtype, is_bias=True)

    hidden = helper.create_variable_for_type_inference(dtype)
    cell = helper.create_variable_for_type_inference(dtype)
    batch_gate = helper.create_variable_for_type_inference(dtype)
    batch_cell_pre_act = helper.create_variable_for_type_inference(dtype)
    inputs = {'Input': input, 'Weight': weight, 'Bias': bias}
    batch_size = input.shape[0]
    if h_0:
        assert h_0.shape == (batch_size, size), \
            'The shape of h0 should be (batch_size, %d)' % size
        inputs['H0'] = h_0
    if c_0:
        assert c_0.shape == (batch_size, size), \
            'The shape of c0 should be (batch_size, %d)' % size
        inputs['C0'] = c_0

    helper.append_op(
        type='lstm',
        inputs=inputs,
        outputs={
            'Hidden': hidden,
            'Cell': cell,
            'BatchGate': batch_gate,
            'BatchCellPreAct': batch_cell_pre_act
        },
        attrs={
            'use_peepholes': use_peepholes,
            'is_reverse': is_reverse,
            'gate_activation': gate_activation,
            'cell_activation': cell_activation,
            'candidate_activation': candidate_activation
        })
    return hidden, cell
Y
Yu Yang 已提交
491 492


P
phlrain 已提交
493 494 495 496 497 498
def lstm(input,
         init_h,
         init_c,
         max_len,
         hidden_size,
         num_layers,
P
phlrain 已提交
499
         dropout_prob=0.0,
P
phlrain 已提交
500 501 502 503 504
         is_bidirec=False,
         is_test=False,
         name=None,
         default_initializer=None,
         seed=-1):
L
liuhongyu 已提交
505
    """
P
phlrain 已提交
506
    If Device is GPU, This op will use cudnn LSTM implementation
L
liuhongyu 已提交
507 508

    A four-gate Long Short-Term Memory network with no peephole connections.
M
minqiyang 已提交
509
    In the forward pass the output ht and cell output ct for a given iteration can be computed from the recurrent input ht-1,
L
liuhongyu 已提交
510 511
    the cell input ct-1 and the previous layer input xt given matrices W, R and biases bW, bR from the following equations:

H
haowang101779990 已提交
512
    .. math::
M
minqiyang 已提交
513 514 515 516 517 518 519

       i_t &= \sigma(W_{ix}x_{t} + W_{ih}h_{t-1} + bx_i + bh_i)

       f_t &= \sigma(W_{fx}x_{t} + W_{fh}h_{t-1} + bx_f + bh_f)

       o_t &= \sigma(W_{ox}x_{t} + W_{oh}h_{t-1} + bx_o + bh_o)

H
haowang101779990 已提交
520
       \\tilde{c_t} &= tanh(W_{cx}x_t + W_{ch}h_{t-1} + bx_c + bh_c)
M
minqiyang 已提交
521 522 523 524

       c_t &= f_t \odot c_{t-1} + i_t \odot \\tilde{c_t}

       h_t &= o_t \odot tanh(c_t)
H
haowang101779990 已提交
525 526

    - $W$ terms denote weight matrices (e.g. $W_{ix}$ is the matrix
P
phlrain 已提交
527 528 529 530 531 532
      of weights from the input gate to the input)
    - The b terms denote bias vectors ($bx_i$ and $bh_i$ are the input gate bias vector).
    - sigmoid is the logistic sigmoid function.
    - $i, f, o$ and $c$ are the input gate, forget gate, output gate,
      and cell activation vectors, respectively, all of which have the same size as
      the cell output activation vector $h$.
H
haowang101779990 已提交
533 534 535
    - The :math:`\odot` is the element-wise product of the vectors.
    - :math:`tanh` is the activation functions.
    - :math:`\\tilde{c_t}` is also called candidate hidden state,
P
phlrain 已提交
536
      which is computed based on the current input and the previous hidden state.
L
liuhongyu 已提交
537

M
minqiyang 已提交
538
    Where sigmoid is the sigmoid operator: :math:`sigmoid(x) = 1 / (1 + e^{-x})` , * represents a point-wise multiplication,
L
liuhongyu 已提交
539 540 541 542 543
    X represensts a matrix multiplication


    Args:
        input (Variable): LSTM input tensor, shape MUST be ( seq_len x batch_size x input_size )
M
minqiyang 已提交
544
        init_h(Variable): The initial hidden state of the LSTM
L
liuhongyu 已提交
545 546 547 548 549
                       This is a tensor with shape ( num_layers x batch_size x hidden_size)
                       if is_bidirec = True, shape should be ( num_layers*2 x batch_size x hidden_size)
        init_c(Variable): The initial cell state of the LSTM.
                       This is a tensor with shape ( num_layers x batch_size x hidden_size )
                       if is_bidirec = True, shape should be ( num_layers*2 x batch_size x hidden_size)
M
minqiyang 已提交
550
        max_len (int): max length of LSTM. the first dim of input tensor CAN NOT greater than max_len
L
liuhongyu 已提交
551 552
        hidden_size (int): hidden size of the LSTM
        num_layers (int): total layers number of the LSTM
P
phlrain 已提交
553 554
        dropout_prob(float|0.0): dropout prob, dropout ONLY work between rnn layers, NOT between time steps
                             There is NO dropout work on rnn output of the last RNN layers
L
liuhongyu 已提交
555 556 557 558 559 560
        is_bidirec (bool): If it is bidirectional
        is_test (bool): If it is in test phrase
        name (str|None): A name for this layer(optional). If set None, the layer
                         will be named automatically.
        default_initializer(Initialize|None): Where use initializer to initialize the Weight
                         If set None, defaule initializer will be used
P
phlrain 已提交
561
        seed(int): Seed for dropout in LSTM, If it's -1, dropout will use random seed
P
phlrain 已提交
562

L
liuhongyu 已提交
563 564

    Returns:
M
minqiyang 已提交
565 566
        rnn_out(Tensor),last_h(Tensor),last_c(Tensor):

H
haowang101779990 已提交
567
                        Three tensors, rnn_out, last_h, last_c:
M
minqiyang 已提交
568

H
haowang101779990 已提交
569 570 571 572
                        - rnn_out is result of LSTM hidden, shape is (seq_len x batch_size x hidden_size) \
                          if is_bidirec set to True, shape will be ( seq_len x batch_sze x hidden_size*2)
                        - last_h is the hidden state of the last step of LSTM \
                          shape is ( num_layers x batch_size x hidden_size ) \
M
minqiyang 已提交
573
                          if is_bidirec set to True, shape will be ( num_layers*2 x batch_size x hidden_size)
H
haowang101779990 已提交
574 575
                        - last_c(Tensor): the cell state of the last step of LSTM \
                          shape is ( num_layers x batch_size x hidden_size ) \
M
minqiyang 已提交
576
                          if is_bidirec set to True, shape will be ( num_layers*2 x batch_size x hidden_size)
L
liuhongyu 已提交
577 578 579 580 581 582 583 584 585 586 587 588 589 590 591


    Examples:
        .. code-block:: python

            input = embedding
            batch_size = 20
            max_len = 100
            dropout_prob = 0.2
            input_size = 100
            hidden_size = 150
            num_layers = 1
            init_hidden1 = layers.fill_constant( [num_layers, batch_size, hidden_size], 'float32', 0.0, stop_grad=False)
            init_cell1 = layers.fill_constant( [num_layers, batch_size, hidden_size], 'float32', 0.0, stop_grad=False)

P
phlrain 已提交
592
            rnn_out, last_h, last_c = layers.lstm( input, init_h, init_c, \
L
liuhongyu 已提交
593 594 595 596 597 598
                    max_len, dropout_prob, input_size, hidden_size, \
                    num_layers)
    """

    helper = LayerHelper('cudnn_lstm', **locals())

P
phlrain 已提交
599 600 601
    dtype = input.dtype
    input_shape = list(input.shape)
    input_size = input_shape[-1]
L
liuhongyu 已提交
602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660
    weight_size = 0
    for i in range(num_layers):
        if i == 0:
            input_weight_size = (input_size * hidden_size) * 4
        else:
            if is_bidirec:
                input_weight_size = (hidden_size * 2 * hidden_size) * 4
            else:
                input_weight_size = (hidden_size * hidden_size) * 4

        hidden_weight_size = (hidden_size * hidden_size) * 4

        if is_bidirec:
            weight_size += (input_weight_size + hidden_weight_size) * 2
            weight_size += hidden_size * 8 * 2
        else:
            weight_size += input_weight_size + hidden_weight_size
            weight_size += hidden_size * 8

    weight = helper.create_parameter(
        attr=helper.param_attr,
        shape=[weight_size],
        dtype=dtype,
        default_initializer=default_initializer)

    out = helper.create_variable_for_type_inference(dtype)
    last_h = helper.create_variable_for_type_inference(dtype)
    last_c = helper.create_variable_for_type_inference(dtype)

    cache = helper.create_variable(
        persistable=True, type=core.VarDesc.VarType.RAW, stop_gradient=True)

    helper.append_op(
        type='cudnn_lstm',
        inputs={
            'Input': input,
            'InitH': init_h,
            'InitC': init_c,
            'W': weight,
            'Cache': cache,
        },
        outputs={
            'Out': out,
            'last_h': last_h,
            'last_c': last_c,
        },
        attrs={
            'max_len': max_len,
            'is_bidirec': is_bidirec,
            'input_size': input_size,
            'hidden_size': hidden_size,
            'num_layers': num_layers,
            'is_test': is_test,
            'dropout_prob': dropout_prob,
            'seed': seed,
        })
    return out, last_h, last_c


Y
Yibing Liu 已提交
661 662 663 664 665 666 667 668 669 670
def dynamic_lstmp(input,
                  size,
                  proj_size,
                  param_attr=None,
                  bias_attr=None,
                  use_peepholes=True,
                  is_reverse=False,
                  gate_activation='sigmoid',
                  cell_activation='tanh',
                  candidate_activation='tanh',
X
xuezhong 已提交
671
                  proj_activation='tanh',
672
                  dtype='float32',
X
xuezhong 已提交
673 674 675 676 677
                  name=None,
                  h_0=None,
                  c_0=None,
                  cell_clip=None,
                  proj_clip=None):
Y
Yibing Liu 已提交
678 679 680
    """
    **Dynamic LSTMP Layer**

681 682 683 684 685 686
    LSTMP (LSTM with recurrent projection) layer has a separate projection
    layer after the LSTM layer, projecting the original hidden state to a
    lower-dimensional one, which is proposed to reduce the number of total
    parameters and furthermore computational complexity for the LSTM,
    espeacially for the case that the size of output units is relative
    large (https://research.google.com/pubs/archive/43905.pdf).
Y
Yibing Liu 已提交
687 688 689 690 691

    The formula is as follows:

    .. math::

692
        i_t & = \sigma(W_{ix}x_{t} + W_{ir}r_{t-1} + W_{ic}c_{t-1} + b_i)
Y
Yibing Liu 已提交
693

694
        f_t & = \sigma(W_{fx}x_{t} + W_{fr}r_{t-1} + W_{fc}c_{t-1} + b_f)
Y
Yibing Liu 已提交
695

696
        \\tilde{c_t} & = act_g(W_{cx}x_t + W_{cr}r_{t-1} + b_c)
Y
Yibing Liu 已提交
697

698
        o_t & = \sigma(W_{ox}x_{t} + W_{or}r_{t-1} + W_{oc}c_t + b_o)
Y
Yibing Liu 已提交
699

700
        c_t & = f_t \odot c_{t-1} + i_t \odot \\tilde{c_t}
Y
Yibing Liu 已提交
701

702
        h_t & = o_t \odot act_h(c_t)
Y
Yibing Liu 已提交
703

704
        r_t & = \overline{act_h}(W_{rh}h_t)
Y
Yibing Liu 已提交
705

Y
Yibing Liu 已提交
706 707 708 709 710 711
    In the above formula:

    * :math:`W`: Denotes weight matrices (e.g. :math:`W_{xi}` is \
          the matrix of weights from the input gate to the input).
    * :math:`W_{ic}`, :math:`W_{fc}`, :math:`W_{oc}`: Diagonal weight \
          matrices for peephole connections. In our implementation, \
712
          we use vectors to reprenset these diagonal weight matrices.
Y
Yibing Liu 已提交
713
    * :math:`b`: Denotes bias vectors (e.g. :math:`b_i` is the input gate \
714
          bias vector).
Y
Yibing Liu 已提交
715 716 717
    * :math:`\sigma`: The activation, such as logistic sigmoid function.
    * :math:`i, f, o` and :math:`c`: The input gate, forget gate, output \
          gate, and cell activation vectors, respectively, all of which have \
718
          the same size as the cell output activation vector :math:`h`.
Y
Yibing Liu 已提交
719
    * :math:`h`: The hidden state.
720
    * :math:`r`: The recurrent projection of the hidden state.
Y
Yibing Liu 已提交
721 722
    * :math:`\\tilde{c_t}`: The candidate hidden state, whose \
          computation is based on the current input and previous hidden state.
723
    * :math:`\odot`: The element-wise product of the vectors.
Y
Yibing Liu 已提交
724
    * :math:`act_g` and :math:`act_h`: The cell input and cell output \
725
          activation functions and `tanh` is usually used for them.
Y
Yibing Liu 已提交
726 727
    * :math:`\overline{act_h}`: The activation function for the projection \
          output, usually using `identity` or same as :math:`act_h`.
Y
Yibing Liu 已提交
728 729 730 731

    Set `use_peepholes` to `False` to disable peephole connection. The formula
    is omitted here, please refer to the paper
    http://www.bioinf.jku.at/publications/older/2604.pdf for details.
732

Y
Yibing Liu 已提交
733 734 735 736 737 738 739 740 741 742 743 744
    Note that these :math:`W_{xi}x_{t}, W_{xf}x_{t}, W_{xc}x_{t}, W_{xo}x_{t}`
    operations on the input :math:`x_{t}` are NOT included in this operator.
    Users can choose to use fully-connected layer before LSTMP layer.

    Args:
        input(Variable): The input of dynamic_lstmp layer, which supports
                         variable-time length input sequence. The underlying
                         tensor in this Variable is a matrix with shape
                         (T X 4D), where T is the total time steps in this
                         mini-batch, D is the hidden size.
        size(int): 4 * hidden size.
        proj_size(int): The size of projection output.
745
        param_attr(ParamAttr|None): The parameter attribute for the learnable
Y
Yibing Liu 已提交
746 747
                               hidden-hidden weight and projection weight.

748 749
                               - Hidden-hidden weight = {:math:`W_{ch}, W_{ih}, \
                                                W_{fh}, W_{oh}`}.
750 751
                               - The shape of hidden-hidden weight is (P x 4D),
                                 where P is the projection size and D the hidden
Y
Yibing Liu 已提交
752 753
                                 size.
                               - Projection weight = {:math:`W_{rh}`}.
754
                               - The shape of projection weight is (D x P).
C
chengduo 已提交
755 756 757 758 759

                               If it is set to None or one attribute of ParamAttr,
                               dynamic_lstm will create ParamAttr as param_attr.
                               If the Initializer of the param_attr is not set, the
                               parameter is initialized with Xavier. Default: None.
760
        bias_attr(ParamAttr|None): The bias attribute for the learnable bias
Y
Yibing Liu 已提交
761 762 763 764 765 766
                              weights, which contains two parts, input-hidden
                              bias weights and peephole connections weights if
                              setting `use_peepholes` to `True`.

                              1. `use_peepholes = False`
                                - Biases = {:math:`b_c, b_i, b_f, b_o`}.
767
                                - The shape is (1 x 4D).
Y
Yibing Liu 已提交
768 769 770
                              2. `use_peepholes = True`
                                - Biases = { :math:`b_c, b_i, b_f, b_o, W_{ic}, \
                                                 W_{fc}, W_{oc}`}.
771
                                - The shape is (1 x 7D).
C
chengduo 已提交
772 773 774 775 776

                              If it is set to None or one attribute of ParamAttr,
                              dynamic_lstm will create ParamAttr as bias_attr.
                              If the Initializer of the bias_attr is not set,
                              the bias is initialized zero. Default: None.
Y
Yibing Liu 已提交
777 778 779 780 781 782 783 784 785
        use_peepholes(bool): Whether to enable diagonal/peephole connections,
                             default `True`.
        is_reverse(bool): Whether to compute reversed LSTM, default `False`.
        gate_activation(str): The activation for input gate, forget gate and
                              output gate. Choices = ["sigmoid", "tanh", "relu",
                              "identity"], default "sigmoid".
        cell_activation(str): The activation for cell output. Choices = ["sigmoid",
                              "tanh", "relu", "identity"], default "tanh".
        candidate_activation(str): The activation for candidate hidden state.
786
                              Choices = ["sigmoid", "tanh", "relu", "identity"],
Y
Yibing Liu 已提交
787 788
                              default "tanh".
        proj_activation(str): The activation for projection output.
789
                              Choices = ["sigmoid", "tanh", "relu", "identity"],
X
xuezhong 已提交
790
                              default "tanh".
Y
Yibing Liu 已提交
791
        dtype(str): Data type. Choices = ["float32", "float64"], default "float32".
792 793
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.
X
xuezhong 已提交
794 795 796 797 798 799 800 801 802 803 804
        h_0(Variable): The initial hidden state is an optional input, default is zero.
                       This is a tensor with shape (N x D), where N is the
                       batch size and D is the projection size.
        c_0(Variable): The initial cell state is an optional input, default is zero.
                       This is a tensor with shape (N x D), where N is the
                       batch size. `h_0` and `c_0` can be NULL but only at the same time.
        cell_clip(float): If provided the cell state is clipped
                             by this value prior to the cell output activation.
        proj_clip(float): If `num_proj > 0` and `proj_clip` is
                            provided, then the projected values are clipped elementwise to within
                            `[-proj_clip, proj_clip]`.
Y
Yibing Liu 已提交
805 806

    Returns:
807 808 809 810
        tuple: A tuple of two output variable: the projection of hidden state, \
               and cell state of LSTMP. The shape of projection is (T x P), \
               for the cell state which is (T x D), and both LoD is the same \
               with the `input`.
Y
Yibing Liu 已提交
811 812

    Examples:
813

Y
Yibing Liu 已提交
814 815
        .. code-block:: python

816 817 818 819
            dict_dim, emb_dim = 128, 64
            data = fluid.layers.data(name='sequence', shape=[1],
                                     dtype='int32', lod_level=1)
            emb = fluid.layers.embedding(input=data, size=[dict_dim, emb_dim])
Y
Yibing Liu 已提交
820
            hidden_dim, proj_dim = 512, 256
821
            fc_out = fluid.layers.fc(input=emb, size=hidden_dim * 4,
Y
Yibing Liu 已提交
822
                                     act=None, bias_attr=None)
823 824 825
            proj_out, _ = fluid.layers.dynamic_lstmp(input=fc_out,
                                                     size=hidden_dim * 4,
                                                     proj_size=proj_dim,
Y
Yibing Liu 已提交
826 827 828 829
                                                     use_peepholes=False,
                                                     is_reverse=True,
                                                     cell_activation="tanh",
                                                     proj_activation="tanh")
Y
Yibing Liu 已提交
830
    """
831

C
chengduo 已提交
832
    assert bias_attr is not False, "bias_attr should not be False in dynamic_lstmp."
Y
Yibing Liu 已提交
833
    helper = LayerHelper('lstmp', **locals())
M
minqiyang 已提交
834
    size = size // 4
Y
Yibing Liu 已提交
835 836 837 838 839 840 841 842 843 844
    weight = helper.create_parameter(
        attr=helper.param_attr, shape=[proj_size, 4 * size], dtype=dtype)
    proj_weight = helper.create_parameter(
        attr=helper.param_attr, shape=[size, proj_size], dtype=dtype)
    bias_size = [1, 7 * size]
    if not use_peepholes:
        bias_size[1] = 4 * size
    bias = helper.create_parameter(
        attr=helper.bias_attr, shape=bias_size, dtype=dtype, is_bias=True)

X
Xin Pan 已提交
845 846 847 848 849 850
    projection = helper.create_variable_for_type_inference(dtype)
    cell = helper.create_variable_for_type_inference(dtype)
    ordered_proj0 = helper.create_variable_for_type_inference(dtype)
    batch_hidden = helper.create_variable_for_type_inference(dtype)
    batch_gate = helper.create_variable_for_type_inference(dtype)
    batch_cell_pre_act = helper.create_variable_for_type_inference(dtype)
851 852 853 854 855 856 857 858 859 860 861 862 863 864 865
    inputs = {
        'Input': input,
        'Weight': weight,
        'ProjWeight': proj_weight,
        'Bias': bias
    }
    batch_size = input.shape[0]
    if h_0:
        assert h_0.shape == (batch_size, proj_size), \
            'The shape of h0 should be (batch_size, %d)' % proj_size
        inputs['H0'] = h_0
    if c_0:
        assert c_0.shape == (batch_size, size), \
            'The shape of c0 should be (batch_size, %d)' % size
        inputs['C0'] = c_0
Y
Yibing Liu 已提交
866

X
xuezhong 已提交
867 868 869 870 871
    if cell_clip:
        assert cell_clip >= 0, "cell_clip should not be negtive."
    if proj_clip:
        assert proj_clip >= 0, "proj_clip should not be negtive."

Y
Yibing Liu 已提交
872 873
    helper.append_op(
        type='lstmp',
874
        inputs=inputs,
Y
Yibing Liu 已提交
875 876 877 878 879 880 881 882 883
        outputs={
            'Projection': projection,
            'Cell': cell,
            'BatchHidden': batch_hidden,
            'BatchGate': batch_gate,
            'BatchCellPreAct': batch_cell_pre_act
        },
        attrs={
            'use_peepholes': use_peepholes,
884 885
            'cell_clip': cell_clip,
            'proj_clip': proj_clip,
Y
Yibing Liu 已提交
886 887 888 889 890 891 892 893 894
            'is_reverse': is_reverse,
            'gate_activation': gate_activation,
            'cell_activation': cell_activation,
            'candidate_activation': candidate_activation,
            'proj_activation': proj_activation
        })
    return projection, cell


G
guosheng 已提交
895 896 897 898 899 900 901
def dynamic_gru(input,
                size,
                param_attr=None,
                bias_attr=None,
                is_reverse=False,
                gate_activation='sigmoid',
                candidate_activation='tanh',
902 903
                h_0=None,
                origin_mode=False):
G
guosheng 已提交
904
    """
905
    **Gated Recurrent Unit (GRU) Layer**
G
guosheng 已提交
906

907 908 909
    if origin_mode is False, then the equation of a gru step is from paper
    `Empirical Evaluation of Gated Recurrent Neural Networks on Sequence
    Modeling <https://arxiv.org/pdf/1412.3555.pdf>`_ .
910

G
guosheng 已提交
911 912 913 914 915 916 917 918 919
    The formula is as follows:

    .. math::

        u_t & = act_g(W_{ux}x_{t} + W_{uh}h_{t-1} + b_u)

        r_t & = act_g(W_{rx}x_{t} + W_{rh}h_{t-1} + b_r)

        \\tilde{h_t} & = act_c(W_{cx}x_{t} + W_{ch}(r_t \odot h_{t-1}) + b_c)
920

G
guosheng 已提交
921
        h_t & = (1-u_t) \odot h_{t-1} + u_t \odot \\tilde{h_t}
922

Q
Qiao Longfei 已提交
923 924 925

    if origin_mode is True then the equation is from paper
    Learning Phrase Representations using RNN Encoder-Decoder for Statistical
926 927 928 929 930 931 932 933 934 935 936 937
    Machine Translation <https://arxiv.org/pdf/1406.1078.pdf>`_

    .. math::

        u_t & = act_g(W_{ux}x_{t} + W_{uh}h_{t-1} + b_u)

        r_t & = act_g(W_{rx}x_{t} + W_{rh}h_{t-1} + b_r)

        \\tilde{h_t} & = act_c(W_{cx}x_{t} + W_{ch}(r_t \odot h_{t-1}) + b_c)

        h_t & = u_t \odot h_{t-1} + (1-u_t) \odot \\tilde{h_t}

G
guosheng 已提交
938
    The :math:`\odot` is the element-wise product of the vectors. :math:`act_g`
939 940
    is the update gate and reset gate activation function and :math:`sigmoid`
    is usually used for it. :math:`act_c` is the activation function for
G
guosheng 已提交
941 942 943 944
    candidate hidden state and :math:`tanh` is usually used for it.

    Note that these :math:`W_{ux}x_{t}, W_{rx}x_{t}, W_{cx}x_{t}` operations on
    the input :math:`x_{t}` are NOT included in this operator. Users can choose
945
    to use fully-connect layer before GRU layer.
G
guosheng 已提交
946 947

    Args:
948 949
        input(Variable): The input of dynamic_gru layer, which supports
            variable-time length input sequence. The underlying tensor in this
G
guosheng 已提交
950
            Variable is a matrix with shape :math:`(T \\times 3D)`, where
951
            :math:`T` is the total time steps in this mini-batch, :math:`D`
G
guosheng 已提交
952 953
            is the hidden size.
        size(int): The dimension of the gru cell.
954
        param_attr(ParamAttr|None): The parameter attribute for the learnable
G
guosheng 已提交
955 956
            hidden-hidden weight matrix. Note:

957
            - The shape of the weight matrix is :math:`(T \\times 3D)`, where
G
guosheng 已提交
958
              :math:`D` is the hidden size.
959
            - All elements in the weight matrix can be divided into two parts.
G
guosheng 已提交
960
              The first part are weights of the update gate and reset gate with
961
              shape :math:`(D \\times 2D)`, and the second part are weights for
G
guosheng 已提交
962
              candidate hidden state with shape :math:`(D \\times D)`.
963 964 965 966 967

            If it is set to None or one attribute of ParamAttr, dynamic_gru will
            create ParamAttr as param_attr. If the Initializer of the param_attr
            is not set, the parameter is initialized with Xavier. Default: None.
        bias_attr (ParamAttr|bool|None): The parameter attribute for the bias
968
            of GRU.Note that the bias with :math:`(1 \\times 3D)` concatenates
969
            the bias in the update gate, reset gate and candidate calculations.
970 971 972
            If it is set to False, no bias will be applied to the update gate,
            reset gate and candidate calculations. If it is set to None or one
            attribute of ParamAttr, dynamic_gru will create ParamAttr as
973 974
            bias_attr. If the Initializer of the bias_attr is not set, the bias
            is initialized zero. Default: None.
975
        is_reverse(bool): Whether to compute reversed GRU, default
G
guosheng 已提交
976 977 978
            :attr:`False`.
        gate_activation(str): The activation for update gate and reset gate.
            Choices = ["sigmoid", "tanh", "relu", "identity"], default "sigmoid".
979
        candidate_activation(str): The activation for candidate hidden state.
G
guosheng 已提交
980
            Choices = ["sigmoid", "tanh", "relu", "identity"], default "tanh".
981 982 983 984
        h_0 (Variable): This is initial hidden state. If not set, default is
            zero. This is a tensor with shape (N x D), where N is the number of
            total time steps of input mini-batch feature and D is the hidden
            size.
G
guosheng 已提交
985 986

    Returns:
G
guosheng 已提交
987
        Variable: The hidden state of GRU. The shape is :math:`(T \\times D)`, \
988
            and sequence length is the same with the input.
989

G
guosheng 已提交
990
    Examples:
991

G
guosheng 已提交
992 993
        .. code-block:: python

994 995 996 997
            dict_dim, emb_dim = 128, 64
            data = fluid.layers.data(name='sequence', shape=[1],
                                     dtype='int32', lod_level=1)
            emb = fluid.layers.embedding(input=data, size=[dict_dim, emb_dim])
G
guosheng 已提交
998
            hidden_dim = 512
999
            x = fluid.layers.fc(input=emb, size=hidden_dim * 3)
T
Tink_Y 已提交
1000
            hidden = fluid.layers.dynamic_gru(input=x, size=hidden_dim)
G
guosheng 已提交
1001 1002 1003 1004 1005 1006 1007 1008 1009
    """

    helper = LayerHelper('gru', **locals())
    dtype = helper.input_dtype()

    weight = helper.create_parameter(
        attr=helper.param_attr, shape=[size, 3 * size], dtype=dtype)
    bias = helper.create_parameter(
        attr=helper.bias_attr, shape=[1, 3 * size], dtype=dtype, is_bias=True)
Y
Yancey 已提交
1010
    batch_size = input.shape[0]
G
guosheng 已提交
1011
    inputs = {'Input': input, 'Weight': weight, 'Bias': bias}
S
sneaxiy 已提交
1012
    if h_0:
G
guosheng 已提交
1013
        assert h_0.shape == (
Y
Yancey 已提交
1014 1015 1016
            batch_size, size
        ), 'The shape of h0 should be(batch_size, %d)' % size
        inputs['H0'] = h_0
G
guosheng 已提交
1017

X
Xin Pan 已提交
1018 1019 1020 1021
    hidden = helper.create_variable_for_type_inference(dtype)
    batch_gate = helper.create_variable_for_type_inference(dtype)
    batch_reset_hidden_prev = helper.create_variable_for_type_inference(dtype)
    batch_hidden = helper.create_variable_for_type_inference(dtype)
G
guosheng 已提交
1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034

    helper.append_op(
        type='gru',
        inputs=inputs,
        outputs={
            'Hidden': hidden,
            'BatchGate': batch_gate,
            'BatchResetHiddenPrev': batch_reset_hidden_prev,
            'BatchHidden': batch_hidden
        },
        attrs={
            'is_reverse': is_reverse,
            'gate_activation': gate_activation,
1035 1036
            'activation': candidate_activation,
            'origin_mode': origin_mode
G
guosheng 已提交
1037 1038 1039 1040
        })
    return hidden


Y
Yu Yang 已提交
1041 1042 1043
def gru_unit(input,
             hidden,
             size,
1044 1045
             param_attr=None,
             bias_attr=None,
Y
Yu Yang 已提交
1046
             activation='tanh',
Q
Qiao Longfei 已提交
1047 1048
             gate_activation='sigmoid',
             origin_mode=False):
Y
Yu Yang 已提交
1049
    """
1050 1051 1052
    **GRU unit layer**

    if origin_mode is True, then the equation of a gru step is from paper
Q
Qiao Longfei 已提交
1053
    `Learning Phrase Representations using RNN Encoder-Decoder for Statistical
1054
    Machine Translation <https://arxiv.org/pdf/1406.1078.pdf>`_
Y
Yu Yang 已提交
1055

1056 1057
        .. math::
            u_t & = actGate(xu_{t} + W_u h_{t-1} + b_u)
Y
Yu Yang 已提交
1058

1059
            r_t & = actGate(xr_{t} + W_r h_{t-1} + b_r)
Y
Yu Yang 已提交
1060

1061
            m_t & = actNode(xm_t + W_c dot(r_t, h_{t-1}) + b_m)
1062

1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077
            h_t & = dot(u_t, h_{t-1}) + dot((1-u_t), m_t)

    if origin_mode is False, then the equation of a gru step is from paper
    `Empirical Evaluation of Gated Recurrent Neural Networks on Sequence
    Modeling <https://arxiv.org/pdf/1412.3555.pdf>`_

        .. math::
            u_t & = actGate(xu_{t} + W_u h_{t-1} + b_u)

            r_t & = actGate(xr_{t} + W_r h_{t-1} + b_r)

            m_t & = actNode(xm_t + W_c dot(r_t, h_{t-1}) + b_m)

            h_t & = dot((1-u_t), h_{t-1}) + dot(u_t, m_t)

1078 1079

    The inputs of gru unit includes :math:`z_t`, :math:`h_{t-1}`. In terms
1080 1081 1082
    of the equation above, the :math:`z_t` is split into 3 parts -
    :math:`xu_t`, :math:`xr_t` and :math:`xm_t`. This means that in order to
    implement a full GRU unit operator for an input, a fully
1083 1084
    connected layer has to be applied, such that :math:`z_t = W_{fc}x_t`.

1085 1086
    The terms :math:`u_t` and :math:`r_t` represent the update and reset gates
    of the GRU cell. Unlike LSTM, GRU has one lesser gate. However, there is
1087 1088 1089
    an intermediate candidate hidden output, which is denoted by :math:`m_t`.
    This layer has three outputs :math:`h_t`, :math:`dot(r_t, h_{t-1})`
    and concatenation of :math:`u_t`, :math:`r_t` and :math:`m_t`.
1090 1091 1092

    Args:
        input (Variable): The fc transformed input value of current step.
1093
        hidden (Variable): The hidden value of gru unit from previous step.
1094
        size (integer): The input dimension value.
1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108
        param_attr(ParamAttr|None): The parameter attribute for the learnable
            hidden-hidden weight matrix. Note:

            - The shape of the weight matrix is :math:`(T \\times 3D)`, where
              :math:`D` is the hidden size.
            - All elements in the weight matrix can be divided into two parts.
              The first part are weights of the update gate and reset gate with
              shape :math:`(D \\times 2D)`, and the second part are weights for
              candidate hidden state with shape :math:`(D \\times D)`.

            If it is set to None or one attribute of ParamAttr, gru_unit will
            create ParamAttr as param_attr. If the Initializer of the param_attr
            is not set, the parameter is initialized with Xavier. Default: None.
        bias_attr (ParamAttr|bool|None): The parameter attribute for the bias
1109
            of GRU.Note that the bias with :math:`(1 \\times 3D)` concatenates
1110
            the bias in the update gate, reset gate and candidate calculations.
1111 1112 1113
            If it is set to False, no bias will be applied to the update gate,
            reset gate and candidate calculations. If it is set to None or one
            attribute of ParamAttr, gru_unit will create ParamAttr as
1114 1115
            bias_attr. If the Initializer of the bias_attr is not set, the bias
            is initialized zero. Default: None.
1116 1117 1118 1119
        activation (string): The activation type for cell (actNode).
                             Default: 'tanh'
        gate_activation (string): The activation type for gates (actGate).
                                  Default: 'sigmoid'
Y
Yu Yang 已提交
1120

1121 1122 1123 1124 1125 1126
    Returns:
        tuple: The hidden value, reset-hidden value and gate values.

    Examples:

        .. code-block:: python
Y
Yu Yang 已提交
1127

1128
             # assuming we have x_t_data and prev_hidden of size=10
1129
             x_t = fluid.layers.fc(input=x_t_data, size=30)
1130 1131
             hidden_val, r_h_val, gate_val = fluid.layers.gru_unit(input=x_t,
                                                    hidden = prev_hidden)
Y
Yu Yang 已提交
1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143

    """
    activation_dict = dict(
        identity=0,
        sigmoid=1,
        tanh=2,
        relu=3, )
    activation = activation_dict[activation]
    gate_activation = activation_dict[gate_activation]

    helper = LayerHelper('gru_unit', **locals())
    dtype = helper.input_dtype()
M
minqiyang 已提交
1144
    size = size // 3
Y
Yu Yang 已提交
1145 1146

    # create weight
1147 1148
    weight = helper.create_parameter(
        attr=helper.param_attr, shape=[size, 3 * size], dtype=dtype)
Y
Yu Yang 已提交
1149

X
Xin Pan 已提交
1150 1151 1152
    gate = helper.create_variable_for_type_inference(dtype)
    reset_hidden_pre = helper.create_variable_for_type_inference(dtype)
    updated_hidden = helper.create_variable_for_type_inference(dtype)
1153
    inputs = {'Input': input, 'HiddenPrev': hidden, 'Weight': weight}
Y
Yu Yang 已提交
1154
    # create bias
1155
    if helper.bias_attr:
Y
Yu Yang 已提交
1156 1157 1158
        bias_size = [1, 3 * size]
        bias = helper.create_parameter(
            attr=helper.bias_attr, shape=bias_size, dtype=dtype, is_bias=True)
1159
        inputs['Bias'] = bias
Y
Yu Yang 已提交
1160 1161 1162

    helper.append_op(
        type='gru_unit',
1163
        inputs=inputs,
Y
Yu Yang 已提交
1164 1165 1166 1167 1168 1169
        outputs={
            'Gate': gate,
            'ResetHiddenPrev': reset_hidden_pre,
            'Hidden': updated_hidden,
        },
        attrs={
1170 1171
            'activation': 2,  # tanh
            'gate_activation': 1,  # sigmoid
Y
Yu Yang 已提交
1172 1173 1174 1175 1176
        })

    return updated_hidden, reset_hidden_pre, gate


Y
yuyang18 已提交
1177
@templatedoc()
1178
def linear_chain_crf(input, label, param_attr=None):
Y
yuyang18 已提交
1179 1180 1181 1182 1183 1184 1185
    """
    Linear Chain CRF.

    ${comment}

    Args:
        input(${emission_type}): ${emission_comment}
D
dzhwinter 已提交
1186
        input(${transition_type}): ${transition_comment}
Y
yuyang18 已提交
1187 1188 1189 1190
        label(${label_type}): ${label_comment}
        param_attr(ParamAttr): The attribute of the learnable parameter.

    Returns:
D
dzhwinter 已提交
1191 1192 1193
        output(${emission_exps_type}): ${emission_exps_comment} \n
        output(${transition_exps_type}): ${transition_exps_comment} \n
        output(${log_likelihood_type}): ${log_likelihood_comment}
Y
yuyang18 已提交
1194 1195

    """
Y
Yu Yang 已提交
1196 1197 1198 1199 1200 1201
    helper = LayerHelper('linear_chain_crf', **locals())
    size = input.shape[1]
    transition = helper.create_parameter(
        attr=helper.param_attr,
        shape=[size + 2, size],
        dtype=helper.input_dtype())
X
Xin Pan 已提交
1202 1203 1204 1205 1206 1207 1208 1209
    alpha = helper.create_variable_for_type_inference(
        dtype=helper.input_dtype())
    emission_exps = helper.create_variable_for_type_inference(
        dtype=helper.input_dtype())
    transition_exps = helper.create_variable_for_type_inference(
        dtype=helper.input_dtype())
    log_likelihood = helper.create_variable_for_type_inference(
        dtype=helper.input_dtype())
Y
Yu Yang 已提交
1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224
    helper.append_op(
        type='linear_chain_crf',
        inputs={"Emission": [input],
                "Transition": transition,
                "Label": label},
        outputs={
            "Alpha": [alpha],
            "EmissionExps": [emission_exps],
            "TransitionExps": transition_exps,
            "LogLikelihood": log_likelihood
        })

    return log_likelihood


W
wopeizl 已提交
1225 1226 1227 1228
@templatedoc()
def crf_decoding(input, param_attr, label=None):
    """
    ${comment}
Y
yi.wu 已提交
1229

W
wopeizl 已提交
1230 1231
    Args:
        input(${emission_type}): ${emission_comment}
Y
yi.wu 已提交
1232

W
wopeizl 已提交
1233
        param_attr(ParamAttr): The parameter attribute for training.
Y
yuyang18 已提交
1234

W
wopeizl 已提交
1235
        label(${label_type}): ${label_comment}
1236

W
wopeizl 已提交
1237 1238
    Returns:
        Variable: ${viterbi_path_comment}
Y
yi.wu 已提交
1239

W
wopeizl 已提交
1240 1241
    Examples:
        .. code-block:: python
Y
yi.wu 已提交
1242

W
wopeizl 已提交
1243 1244 1245 1246 1247 1248 1249 1250 1251 1252
           crf_decode = layers.crf_decoding(
                input=hidden, param_attr=ParamAttr(name="crfw"))
    """
    helper = LayerHelper('crf_decoding', **locals())
    transition = helper.get_parameter(param_attr.name)
    viterbi_path = helper.create_variable_for_type_inference(
        dtype=helper.input_dtype())
    helper.append_op(
        type='crf_decoding',
        inputs={"Emission": [input],
Y
Yu Yang 已提交
1253
                "Transition": transition,
W
wopeizl 已提交
1254 1255
                "Label": label},
        outputs={"ViterbiPath": [viterbi_path]})
Y
Yu Yang 已提交
1256

W
wopeizl 已提交
1257
    return viterbi_path
Y
Yu Yang 已提交
1258 1259


Y
yi.wu 已提交
1260
@templatedoc()
F
fengjiayi 已提交
1261
def cos_sim(X, Y):
Y
Yu Yang 已提交
1262
    """
Y
yi.wu 已提交
1263 1264 1265
    ${comment}

    Args:
1266 1267
        X (Variable): ${x_comment}.
        Y (Variable): ${y_comment}.
F
fengjiayi 已提交
1268

Y
yi.wu 已提交
1269
    Returns:
1270
        Variable: the output of cosine(X, Y).
Y
Yu Yang 已提交
1271
    """
F
fengjiayi 已提交
1272
    helper = LayerHelper('cos_sim', **locals())
X
Xin Pan 已提交
1273 1274 1275
    out = helper.create_variable_for_type_inference(dtype=X.dtype)
    xnorm = helper.create_variable_for_type_inference(dtype=X.dtype)
    ynorm = helper.create_variable_for_type_inference(dtype=X.dtype)
Y
Yu Yang 已提交
1276 1277 1278 1279 1280 1281 1282 1283 1284 1285
    helper.append_op(
        type='cos_sim',
        inputs={'X': [X],
                'Y': [Y]},
        outputs={'Out': [out],
                 'XNorm': [xnorm],
                 'YNorm': [ynorm]})
    return out


P
phlrain 已提交
1286 1287 1288 1289 1290
def dropout(x,
            dropout_prob,
            is_test=False,
            seed=None,
            name=None,
P
phlrain 已提交
1291
            dropout_implementation="downgrade_in_infer"):
1292 1293 1294 1295 1296
    """
    Computes dropout.

    Drop or keep each element of `x` independently. Dropout is a regularization
    technique for reducing overfitting by preventing neuron co-adaption during
1297
    training. The dropout operator randomly sets (according to the given dropout
1298 1299 1300
    probability) the outputs of some units to zero, while others are remain
    unchanged.

H
haowang101779990 已提交
1301 1302
    dropout op can be removed from the program to make the program more efficient.

1303
    Args:
1304 1305
        x (Variable): The input tensor variable.
        dropout_prob (float): Probability of setting units to zero.
1306 1307 1308 1309 1310 1311 1312
        is_test (bool): A flag indicating whether it is in test phrase or not.
        seed (int): A Python integer used to create random seeds. If this
                    parameter is set to None, a random seed is used.
                    NOTE: If an integer seed is given, always the same output
                    units will be dropped. DO NOT use a fixed seed in training.
        name (str|None): A name for this layer(optional). If set None, the layer
                         will be named automatically.
H
haowang101779990 已提交
1313 1314
        dropout_implementation(string): ['downgrade_in_infer'(default)|'upscale_in_train']

P
phlrain 已提交
1315
                                        1. downgrade_in_infer(default), downgrade the outcome at inference
H
haowang101779990 已提交
1316 1317 1318 1319 1320 1321

                                           - train: out = input * mask
                                           - inference: out = input * dropout_prob

                                           (mask is a tensor same shape with input, value is 0 or 1
                                           ratio of 0 is dropout_prob)
P
phlrain 已提交
1322
                                        2. upscale_in_train, upscale the outcome at training time
1323

H
haowang101779990 已提交
1324 1325
                                           - train: out = input * mask / ( 1.0 - dropout_prob )
                                           - inference: out = input
P
phlrain 已提交
1326

H
haowang101779990 已提交
1327 1328
                                           (mask is a tensor same shape with input, value is 0 or 1
                                           ratio of 0 is dropout_prob)
1329

M
minqiyang 已提交
1330

1331
    Returns:
1332
        Variable: A tensor variable is the shape with `x`.
1333 1334

    Examples:
1335

1336 1337
        .. code-block:: python

1338 1339
            x = fluid.layers.data(name="data", shape=[32, 32], dtype="float32")
            droped = fluid.layers.dropout(x, dropout_prob=0.5)
1340 1341
    """

F
fengjiayi 已提交
1342
    helper = LayerHelper('dropout', **locals())
X
Xin Pan 已提交
1343 1344 1345
    out = helper.create_variable_for_type_inference(dtype=x.dtype)
    mask = helper.create_variable_for_type_inference(
        dtype=x.dtype, stop_gradient=True)
C
chengduo 已提交
1346 1347 1348 1349

    if (seed is None or seed == 0) and helper.main_program.random_seed != 0:
        seed = helper.main_program.random_seed

1350 1351 1352 1353 1354
    helper.append_op(
        type='dropout',
        inputs={'X': [x]},
        outputs={'Out': [out],
                 'Mask': [mask]},
1355 1356 1357 1358
        attrs={
            'dropout_prob': dropout_prob,
            'is_test': is_test,
            'fix_seed': seed is not None,
P
phlrain 已提交
1359 1360
            'seed': seed if seed is not None else 0,
            'dropout_implementation': dropout_implementation,
1361
        })
1362 1363 1364
    return out


J
jerrywgz 已提交
1365
def cross_entropy(input, label, soft_label=False, ignore_index=kIgnoreIndex):
Y
Yu Yang 已提交
1366
    """
Y
Yibing Liu 已提交
1367 1368
    **Cross Entropy Layer**

1369 1370 1371
    This layer computes the cross entropy between `input` and `label`. It
    supports both standard cross-entropy and soft-label cross-entropy loss
    computation.
Y
Yibing Liu 已提交
1372 1373

    1) One-hot cross-entropy:
F
fengjiayi 已提交
1374
        `soft_label = False`, `Label[i, 0]` indicates the class index for sample i:
Y
yangyaming 已提交
1375

Y
Yibing Liu 已提交
1376
        .. math::
Y
yangyaming 已提交
1377

Y
Yibing Liu 已提交
1378 1379 1380
            Y[i] = -\log(X[i, Label[i]])

    2) Soft-label cross-entropy:
F
fengjiayi 已提交
1381 1382
        `soft_label = True`, `Label[i, j]` indicates the soft label of class j
        for sample i:
Y
Yibing Liu 已提交
1383 1384 1385 1386 1387

        .. math::

            Y[i] = \sum_j{-Label[i, j] * log(X[i, j])}

Y
Yibing Liu 已提交
1388
       Please make sure that in this case the summation of each row of `label`
Y
Yibing Liu 已提交
1389 1390 1391
       equals one.

    3) One-hot cross-entropy with vecterized `label`:
F
fengjiayi 已提交
1392 1393
         As a special case of 2), when each row of 'label' has only one
         non-zero element which is equal to 1, soft-label cross-entropy degenerates
Y
Yibing Liu 已提交
1394
         to a one-hot cross-entropy with one-hot label representation.
Y
yangyaming 已提交
1395

Y
Yibing Liu 已提交
1396
    Args:
Y
yangyaming 已提交
1397
        input (Variable|list):  a 2-D tensor with shape [N x D], where N is the
1398 1399 1400 1401
                                batch size and D is the number of classes. This
                                input is a probability computed by the previous
                                operator, which is almost always the result of
                                a softmax operator.
Y
yangyaming 已提交
1402
        label (Variable|list): the ground truth which is a 2-D tensor. When
1403 1404 1405 1406
                               `soft_label` is set to `False`, `label` is a
                               tensor<int64> with shape [N x 1]. When
                               `soft_label` is set to `True`, `label` is a
                               tensor<float/double> with shape [N x D].
F
fengjiayi 已提交
1407
        soft_label (bool): a flag indicating whether to
1408
                                           interpretate the given labels as soft
1409
                                           labels. Default: `False`.
M
minqiyang 已提交
1410 1411
        ignore_index (int): Specifies a target value that is ignored and does
                            not contribute to the input gradient. Only valid
J
jerrywgz 已提交
1412
                            if soft_label is set to False. Default: kIgnoreIndex
Y
Yibing Liu 已提交
1413 1414 1415 1416 1417

    Returns:
         A 2-D tensor with shape [N x 1], the cross entropy loss.

    Raises:
H
haowang101779990 已提交
1418 1419 1420
         ValueError:

                      1. the 1st dimension of ``input`` and ``label`` are not equal.
M
minqiyang 已提交
1421

H
haowang101779990 已提交
1422 1423
                      2. when ``soft_label == True``, and the 2nd dimension of
                         ``input`` and ``label`` are not equal.
M
minqiyang 已提交
1424

H
haowang101779990 已提交
1425 1426
                      3. when ``soft_label == False``, and the 2nd dimension of
                         ``label`` is not 1.
Y
Yibing Liu 已提交
1427 1428 1429 1430 1431 1432

    Examples:
        .. code-block:: python

          predict = fluid.layers.fc(input=net, size=classdim, act='softmax')
          cost = fluid.layers.cross_entropy(input=predict, label=label)
Y
Yu Yang 已提交
1433
    """
F
fengjiayi 已提交
1434
    helper = LayerHelper('cross_entropy', **locals())
X
Xin Pan 已提交
1435
    out = helper.create_variable_for_type_inference(dtype=input.dtype)
Y
Yu Yang 已提交
1436 1437 1438 1439 1440
    helper.append_op(
        type='cross_entropy',
        inputs={'X': [input],
                'Label': [label]},
        outputs={'Y': [out]},
1441 1442
        attrs={"soft_label": soft_label,
               "ignore_index": ignore_index})
Y
Yu Yang 已提交
1443 1444 1445
    return out


F
frankwhzhang 已提交
1446
def bpr_loss(input, label, name=None):
F
frankwhzhang 已提交
1447 1448 1449
    """
    Bayesian Personalized Ranking Loss Operator.

1450
    This operator belongs to pairwise ranking loss. Label is the desired item.
F
frankwhzhang 已提交
1451 1452 1453 1454 1455 1456
    The loss at a given point in one session is defined as:
    $Y[i] = -\frac{1}{N_{i}-1} * \sum_{0\le j<N_{i},~ j\neq Label[i]}\log(\sigma(X[i, Label[i]]-X[i, j]))$

    Learn more details by reading paper <session-based recommendations with recurrent
    neural networks>(https://arxiv.org/abs/1511.06939)

1457 1458 1459 1460 1461 1462
    Args:
        input (Variable|list):  a 2-D tensor with shape [N x D], where N is the
                                batch size and D is the number of classes.
                                This input is not probability but logits.
        label (Variable|list):  the ground truth which is a 2-D tensor.  `label`
                                is a tensor<int64> with shape [N x 1].
F
frankwhzhang 已提交
1463 1464
        name (str|None):        A name for this layer(optional). If set None, the
                                layer will be named automatically. Default: None.
1465 1466 1467
    Returns:
        A 2-D tensor with shape [N x 1], the bpr loss.

F
frankwhzhang 已提交
1468 1469 1470
    Examples:
        .. code-block:: python

1471
          cost = fluid.layers.bpr_loss(input=predict, label=label)
F
frankwhzhang 已提交
1472
    """
1473 1474 1475 1476 1477 1478

    helper = LayerHelper('bpr_loss', **locals())
    out = helper.create_variable_for_type_inference(dtype=input.dtype)
    helper.append_op(
        type='bpr_loss',
        inputs={'X': [input],
1479
                'Label': [label]},
1480 1481 1482 1483
        outputs={'Y': [out]})
    return out


F
fengjiayi 已提交
1484
def square_error_cost(input, label):
Y
Yu Yang 已提交
1485
    """
1486 1487
    **Square error cost layer**

1488 1489
    This layer accepts input predictions and target label and returns the
    squared error cost.
Y
ying 已提交
1490

1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503
    For predictions, :math:`X`, and target labels, :math:`Y`, the equation is:

    .. math::

        Out = (X - Y)^2

    In the above equation:

        * :math:`X`: Input predictions, a tensor.
        * :math:`Y`: Input labels, a tensor.
        * :math:`Out`: Output value, same shape with :math:`X`.

    Args:
1504 1505
        input (Variable): Input tensor, has predictions.
        label (Variable): Label tensor, has target labels.
1506 1507

    Returns:
G
guosheng 已提交
1508
        Variable: The tensor variable storing the element-wise squared error \
1509
                  difference of input and label.
1510 1511 1512 1513 1514 1515 1516 1517

    Examples:
        .. code-block:: python

          y = layers.data(name='y', shape=[1], dtype='float32')
          y_predict = layers.data(name='y_predict', shape=[1], dtype='float32')
          cost = layers.square_error_cost(input=y_predict, label=y)

Y
Yu Yang 已提交
1518
    """
F
fengjiayi 已提交
1519
    helper = LayerHelper('square_error_cost', **locals())
X
Xin Pan 已提交
1520
    minus_out = helper.create_variable_for_type_inference(dtype=input.dtype)
Y
Yu Yang 已提交
1521 1522 1523 1524 1525 1526
    helper.append_op(
        type='elementwise_sub',
        inputs={'X': [input],
                'Y': [label]},
        outputs={'Out': [minus_out]})

X
Xin Pan 已提交
1527
    square_out = helper.create_variable_for_type_inference(dtype=input.dtype)
Y
Yu Yang 已提交
1528
    helper.append_op(
F
fengjiayi 已提交
1529 1530
        type='square', inputs={'X': [minus_out]},
        outputs={'Out': [square_out]})
Y
Yu Yang 已提交
1531 1532 1533
    return square_out


Y
yi.wu 已提交
1534
@templatedoc()
Y
Yu Yang 已提交
1535 1536 1537 1538
def chunk_eval(input,
               label,
               chunk_scheme,
               num_chunk_types,
F
fengjiayi 已提交
1539
               excluded_chunk_types=None):
Y
Yu Yang 已提交
1540
    """
Y
yi.wu 已提交
1541
    **Chunk Evaluator**
Y
yi.wu 已提交
1542

Y
yangyaming 已提交
1543
    This function computes and outputs the precision, recall and
1544
    F1-score of chunk detection.
Y
yi.wu 已提交
1545

M
minqiyang 已提交
1546
    For some basics of chunking, please refer to
H
haowang101779990 已提交
1547
    `Chunking with Support Vector Machines <https://aclanthology.info/pdf/N/N01/N01-1025.pdf>`_ .
Y
yi.wu 已提交
1548 1549 1550 1551 1552 1553

    ChunkEvalOp computes the precision, recall, and F1-score of chunk detection,
    and supports IOB, IOE, IOBES and IO (also known as plain) tagging schemes.
    Here is a NER example of labeling for these tagging schemes:

    .. code-block:: python
1554

Y
yi.wu 已提交
1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579
       ====== ====== ======  =====  ==  ============   =====  ===== =====  ==  =========
              Li     Ming    works  at  Agricultural   Bank   of    China  in  Beijing.
       ====== ====== ======  =====  ==  ============   =====  ===== =====  ==  =========
       IO     I-PER  I-PER   O      O   I-ORG          I-ORG  I-ORG I-ORG  O   I-LOC
       IOB    B-PER  I-PER   O      O   B-ORG          I-ORG  I-ORG I-ORG  O   B-LOC
       IOE    I-PER  E-PER   O      O   I-ORG          I-ORG  I-ORG E-ORG  O   E-LOC
       IOBES  B-PER  E-PER   O      O   I-ORG          I-ORG  I-ORG E-ORG  O   S-LOC
       ====== ====== ======  =====  ==  ============   =====  ===== =====  ==  =========

    There are three chunk types(named entity types) including PER(person), ORG(organization)
    and LOC(LOCATION), and we can see that the labels have the form <tag type>-<chunk type>.

    Since the calculations actually use label ids rather than labels, extra attention
    should be paid when mapping labels to ids to make CheckEvalOp work. The key point
    is that the listed equations are satisfied by ids.

    .. code-block:: python

       tag_type = label % num_tag_type
       chunk_type = label / num_tag_type

    where `num_tag_type` is the num of tag types in the tagging scheme, `num_chunk_type`
    is the num of chunk types, and `tag_type` get its value from the following table.

    .. code-block:: python
1580

Y
yi.wu 已提交
1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604
       Scheme Begin Inside End   Single
        plain   0     -      -     -
        IOB     0     1      -     -
        IOE     -     0      1     -
        IOBES   0     1      2     3

    Still use NER as example, assuming the tagging scheme is IOB while chunk types are ORG,
    PER and LOC. To satisfy the above equations, the label map can be like this:

    .. code-block:: python

       B-ORG  0
       I-ORG  1
       B-PER  2
       I-PER  3
       B-LOC  4
       I-LOC  5
       O      6

    It's not hard to verify the equations noting that the num of chunk types
    is 3 and the num of tag types in IOB scheme is 2. For example, the label
    id of I-LOC is 5, the tag type id of I-LOC is 1, and the chunk type id of
    I-LOC is 2, which consistent with the results from the equations.

Y
yi.wu 已提交
1605
    Args:
1606 1607 1608 1609 1610
        input (Variable): prediction output of the network.
        label (Variable): label of the test data set.
        chunk_scheme (str): ${chunk_scheme_comment}
        num_chunk_types (int): ${num_chunk_types_comment}
        excluded_chunk_types (list): ${excluded_chunk_types_comment}
F
fengjiayi 已提交
1611

Y
yi.wu 已提交
1612
    Returns:
Y
update  
yi.wu 已提交
1613 1614 1615
        tuple: tuple containing: precision, recall, f1_score,
        num_infer_chunks, num_label_chunks,
        num_correct_chunks
1616

Y
yi.wu 已提交
1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628
    Examples:
        .. code-block:: python

            crf = fluid.layers.linear_chain_crf(
                input=hidden, label=label, param_attr=ParamAttr(name="crfw"))
            crf_decode = fluid.layers.crf_decoding(
                input=hidden, param_attr=ParamAttr(name="crfw"))
            fluid.layers.chunk_eval(
                input=crf_decode,
                label=label,
                chunk_scheme="IOB",
                num_chunk_types=(label_dict_len - 1) / 2)
Y
Yu Yang 已提交
1629
    """
F
fengjiayi 已提交
1630
    helper = LayerHelper("chunk_eval", **locals())
Y
Yu Yang 已提交
1631 1632

    # prepare output
X
Xin Pan 已提交
1633 1634 1635 1636 1637 1638 1639
    precision = helper.create_variable_for_type_inference(dtype="float32")
    recall = helper.create_variable_for_type_inference(dtype="float32")
    f1_score = helper.create_variable_for_type_inference(dtype="float32")
    num_infer_chunks = helper.create_variable_for_type_inference(dtype="int64")
    num_label_chunks = helper.create_variable_for_type_inference(dtype="int64")
    num_correct_chunks = helper.create_variable_for_type_inference(
        dtype="int64")
Y
Yu Yang 已提交
1640 1641 1642 1643 1644 1645 1646 1647

    helper.append_op(
        type="chunk_eval",
        inputs={"Inference": [input],
                "Label": [label]},
        outputs={
            "Precision": [precision],
            "Recall": [recall],
1648 1649 1650 1651
            "F1-Score": [f1_score],
            "NumInferChunks": [num_infer_chunks],
            "NumLabelChunks": [num_label_chunks],
            "NumCorrectChunks": [num_correct_chunks]
Y
Yu Yang 已提交
1652 1653 1654
        },
        attrs={
            "num_chunk_types": num_chunk_types,
G
guosheng 已提交
1655 1656
            "chunk_scheme": chunk_scheme,
            "excluded_chunk_types": excluded_chunk_types or []
Y
Yu Yang 已提交
1657
        })
1658 1659
    return (precision, recall, f1_score, num_infer_chunks, num_label_chunks,
            num_correct_chunks)
Y
Yu Yang 已提交
1660 1661


1662
@templatedoc()
Y
Yu Yang 已提交
1663 1664 1665 1666 1667 1668 1669
def sequence_conv(input,
                  num_filters,
                  filter_size=3,
                  filter_stride=1,
                  padding=None,
                  bias_attr=None,
                  param_attr=None,
C
chengduo 已提交
1670 1671
                  act=None,
                  name=None):
Y
Yu Yang 已提交
1672 1673 1674 1675
    """
    This function creates the op for sequence_conv, using the inputs and
    other convolutional configurations for the filters and stride as given
    in the input parameters to the function.
1676 1677 1678 1679 1680 1681 1682

    Args:
        input (Variable): ${x_comment}
        num_filters (int): number of filters.
        filter_size (int): the filter size (H and W).
        filter_stride (int): stride of the filter.
        padding (bool): if True, add paddings.
C
chengduo 已提交
1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695
        bias_attr (ParamAttr|bool|None): The parameter attribute for the bias of sequence_conv.
            If it is set to False, no bias will be added to the output units.
            If it is set to None or one attribute of ParamAttr, sequence_conv
            will create ParamAttr as bias_attr. If the Initializer of the bias_attr
            is not set, the bias is initialized zero. Default: None.
        param_attr (ParamAttr|None): The parameter attribute for learnable parameters/weights
            of sequence_conv. If it is set to None or one attribute of ParamAttr, sequence_conv
            will create ParamAttr as param_attr. If the Initializer of the param_attr
            is not set, the parameter is initialized with Xavier. Default: None.
        act (str): Activation type, if it is set to None, activation is not appended.
            Default: None.
        name (str|None): A name for this layer(optional). If set None, the layer
            will be named automatically. Default: None.
F
fengjiayi 已提交
1696

1697 1698
    Returns:
        Variable: output of sequence_conv
Y
Yu Yang 已提交
1699 1700 1701 1702 1703 1704 1705
    """

    helper = LayerHelper('sequence_conv', **locals())
    dtype = helper.input_dtype()
    filter_shape = [filter_size * input.shape[1], num_filters]
    filter_param = helper.create_parameter(
        attr=helper.param_attr, shape=filter_shape, dtype=dtype)
X
Xin Pan 已提交
1706
    pre_bias = helper.create_variable_for_type_inference(dtype)
Y
Yu Yang 已提交
1707 1708 1709 1710 1711 1712 1713 1714 1715 1716

    helper.append_op(
        type='sequence_conv',
        inputs={
            'X': [input],
            'Filter': [filter_param],
        },
        outputs={"Out": pre_bias},
        attrs={
            'contextStride': filter_stride,
M
minqiyang 已提交
1717
            'contextStart': -int(filter_size // 2),
Y
Yu Yang 已提交
1718 1719 1720 1721 1722 1723
            'contextLength': filter_size
        })
    pre_act = helper.append_bias_op(pre_bias)
    return helper.append_activation(pre_act)


C
chengduo 已提交
1724
def sequence_softmax(input, use_cudnn=False, name=None):
1725 1726 1727
    """
    This function computes the softmax activation among all time-steps for each
    sequence. The dimension of each time-step should be 1. Thus, the shape of
1728
    input Tensor can be either :math:`[N, 1]` or :math:`[N]`, where :math:`N`
1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744
    is the sum of the length of all sequences.

    For i-th sequence in a mini-batch:

    .. math::

        Out(X[lod[i]:lod[i+1]], :) = \\frac{\exp(X[lod[i]:lod[i+1], :])}{\sum(\exp(X[lod[i]:lod[i+1], :]))}

    For example, for a mini-batch of 3 sequences with variable-length,
    each containing 2, 3, 2 time-steps, the lod of which is [0, 2, 5, 7],
    then softmax will be computed among :math:`X[0:2, :]`, :math:`X[2:5, :]`,
    :math:`X[5:7, :]`, and :math:`N` turns out to be 7.

    Args:
        input (Variable): The input variable which is a LoDTensor.
        use_cudnn (bool): Use cudnn kernel or not, it is valid only when the cudnn \
C
chengduo 已提交
1745 1746 1747
            library is installed. Default: False.
        name (str|None): A name for this layer(optional). If set None, the layer
            will be named automatically. Default: None.
1748

1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759
    Returns:
        Variable: output of sequence_softmax

    Examples:

        .. code-block:: python

             x = fluid.layers.data(name='x', shape=[7, 1],
                              dtype='float32', lod_level=1)
             x_sequence_softmax = fluid.layers.sequence_softmax(input=x)
    """
1760 1761
    helper = LayerHelper('sequence_softmax', **locals())
    dtype = helper.input_dtype()
X
Xin Pan 已提交
1762
    softmax_out = helper.create_variable_for_type_inference(dtype)
1763 1764 1765 1766 1767 1768 1769 1770
    helper.append_op(
        type="sequence_softmax",
        inputs={"X": input},
        outputs={"Out": softmax_out},
        attrs={"use_cudnn": use_cudnn})
    return softmax_out


1771
def softmax(input, use_cudnn=False, name=None):
Q
qiaolongfei 已提交
1772
    """
1773
    The input of the softmax operator is a tensor of any rank. The output tensor
F
fengjiayi 已提交
1774
    has the same shape as the input.
Q
qiaolongfei 已提交
1775

1776 1777 1778 1779 1780 1781
    The input tensor will first be logically flattened to a 2-D matrix. The matrix's
    second dimension(row length) is as same as the last dimension of the input
    tensor, and the first dimension(column length) is the product of all other
    dimensions of the input tensor. For each row of the matrix, the softmax operator
    squashes the K-dimensional(K is the width of the matrix, which is also the size
    of the input tensor's last dimension) vector of arbitrary real values to a
F
fengjiayi 已提交
1782
    K-dimensional vector of real values in the range [0, 1] that add up to 1.
Q
qiaolongfei 已提交
1783 1784 1785 1786 1787 1788 1789

    It computes the exponential of the given dimension and the sum of exponential
    values of all the other dimensions in the K-dimensional vector input.
    Then the ratio of the exponential of the given dimension and the sum of
    exponential values of all the other dimensions is the output of the softmax
    operator.

F
fengjiayi 已提交
1790
    For each row :math:`i` and each column :math:`j` in the matrix, we have:
Q
qiaolongfei 已提交
1791 1792 1793 1794 1795 1796 1797 1798

    .. math::

        Out[i, j] = \\frac{\exp(X[i, j])}{\sum_j(exp(X[i, j])}

    Args:
        input (Variable): The input variable.
        use_cudnn (bool): Use cudnn kernel or not, it is valid only when the cudnn \
C
chengduo 已提交
1799 1800 1801
            library is installed.
        name (str|None): A name for this layer(optional). If set None, the layer
            will be named automatically. Default: None.
Q
qiaolongfei 已提交
1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813

    Returns:
        Variable: output of softmax

    Examples:

        .. code-block:: python

             fc = fluid.layers.fc(input=x, size=10)
             softmax = fluid.layers.softmax(input=fc)

    """
1814 1815
    helper = LayerHelper('softmax', **locals())
    dtype = helper.input_dtype()
X
Xin Pan 已提交
1816
    softmax_out = helper.create_variable_for_type_inference(dtype)
1817 1818 1819 1820 1821 1822 1823 1824
    helper.append_op(
        type="softmax",
        inputs={"X": input},
        outputs={"Out": softmax_out},
        attrs={"use_cudnn": use_cudnn})
    return softmax_out


Y
Yu Yang 已提交
1825 1826 1827
def conv2d(input,
           num_filters,
           filter_size,
C
chengduoZH 已提交
1828 1829
           stride=1,
           padding=0,
1830
           dilation=1,
Y
Yu Yang 已提交
1831 1832 1833
           groups=None,
           param_attr=None,
           bias_attr=None,
C
chengduoZH 已提交
1834
           use_cudnn=True,
1835 1836
           act=None,
           name=None):
Y
Yu Yang 已提交
1837
    """
C
chengduoZH 已提交
1838
    The convolution2D layer calculates the output based on the input, filter
T
tensor-tang 已提交
1839 1840
    and strides, paddings, dilations, groups parameters. Input and
    Output are in NCHW format, where N is batch size, C is the number of
1841
    channels, H is the height of the feature, and W is the width of the feature.
T
tensor-tang 已提交
1842 1843 1844 1845 1846 1847 1848
    Filter is in MCHW format, where M is the number of output image channels,
    C is the number of input image channels, H is the height of the filter,
    and W is the width of the filter. If the groups is greater than 1,
    C will equal the number of input image channels divided by the groups.
    Please refer to UFLDL's `convolution
    <http://ufldl.stanford.edu/tutorial/supervised/FeatureExtractionUsingConvolution/>`_
    for more detials.
1849 1850 1851
    If bias attribution and activation type are provided, bias is added to the
    output of the convolution, and the corresponding activation function is
    applied to the final result.
C
chengduoZH 已提交
1852

1853
    For each input :math:`X`, the equation is:
C
refine  
chengduoZH 已提交
1854

C
chengduoZH 已提交
1855 1856
    .. math::

C
refine  
chengduoZH 已提交
1857
        Out = \sigma (W \\ast X + b)
C
chengduoZH 已提交
1858

T
tensor-tang 已提交
1859
    Where:
C
chengduoZH 已提交
1860

1861 1862 1863 1864 1865
    * :math:`X`: Input value, a tensor with NCHW format.
    * :math:`W`: Filter value, a tensor with MCHW format.
    * :math:`\\ast`: Convolution operation.
    * :math:`b`: Bias value, a 2-D tensor with shape [M, 1].
    * :math:`\\sigma`: Activation function.
T
tensor-tang 已提交
1866
    * :math:`Out`: Output value, the shape of :math:`Out` and :math:`X` may be different.
C
chengduoZH 已提交
1867 1868 1869

    Example:

1870 1871
        - Input:

W
weixing02 已提交
1872
          Input shape: :math:`(N, C_{in}, H_{in}, W_{in})`
C
refine  
chengduoZH 已提交
1873

W
weixing02 已提交
1874
          Filter shape: :math:`(C_{out}, C_{in}, H_f, W_f)`
C
refine  
chengduoZH 已提交
1875

1876
        - Output:
T
tensor-tang 已提交
1877

W
weixing02 已提交
1878
          Output shape: :math:`(N, C_{out}, H_{out}, W_{out})`
C
refine  
chengduoZH 已提交
1879

C
chengduoZH 已提交
1880
        Where
1881 1882

        .. math::
C
chengduoZH 已提交
1883

W
weixing02 已提交
1884 1885
            H_{out}&= \\frac{(H_{in} + 2 * paddings[0] - (dilations[0] * (H_f - 1) + 1))}{strides[0]} + 1 \\\\
            W_{out}&= \\frac{(W_{in} + 2 * paddings[1] - (dilations[1] * (W_f - 1) + 1))}{strides[1]} + 1
C
chengduoZH 已提交
1886 1887

    Args:
1888
        input (Variable): The input image with [N, C, H, W] format.
T
tensor-tang 已提交
1889
        num_filters(int): The number of filter. It is as same as the output
1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906
            image channel.
        filter_size (int|tuple|None): The filter size. If filter_size is a tuple,
            it must contain two integers, (filter_size_H, filter_size_W).
            Otherwise, the filter will be a square.
        stride (int|tuple): The stride size. If stride is a tuple, it must
            contain two integers, (stride_H, stride_W). Otherwise, the
            stride_H = stride_W = stride. Default: stride = 1.
        padding (int|tuple): The padding size. If padding is a tuple, it must
            contain two integers, (padding_H, padding_W). Otherwise, the
            padding_H = padding_W = padding. Default: padding = 0.
        dilation (int|tuple): The dilation size. If dilation is a tuple, it must
            contain two integers, (dilation_H, dilation_W). Otherwise, the
            dilation_H = dilation_W = dilation. Default: dilation = 1.
        groups (int): The groups number of the Conv2d Layer. According to grouped
            convolution in Alex Krizhevsky's Deep CNN paper: when group=2,
            the first half of the filters is only connected to the first half
            of the input channels, while the second half of the filters is only
C
chengduo 已提交
1907 1908 1909 1910 1911
            connected to the second half of the input channels. Default: groups=1.
        param_attr (ParamAttr|None): The parameter attribute for learnable parameters/weights
            of conv2d. If it is set to None or one attribute of ParamAttr, conv2d
            will create ParamAttr as param_attr. If the Initializer of the param_attr
            is not set, the parameter is initialized with :math:`Normal(0.0, std)`,
H
haowang101779990 已提交
1912
            and the :math:`std` is :math:`(\\frac{2.0 }{filter\_elem\_num})^{0.5}`. Default: None.
C
chengduo 已提交
1913 1914 1915 1916 1917
        bias_attr (ParamAttr|bool|None): The parameter attribute for the bias of conv2d.
            If it is set to False, no bias will be added to the output units.
            If it is set to None or one attribute of ParamAttr, conv2d
            will create ParamAttr as bias_attr. If the Initializer of the bias_attr
            is not set, the bias is initialized zero. Default: None.
1918 1919
        use_cudnn (bool): Use cudnn kernel or not, it is valid only when the cudnn
            library is installed. Default: True
C
chengduo 已提交
1920 1921
        act (str): Activation type, if it is set to None, activation is not appended.
            Default: None
1922
        name (str|None): A name for this layer(optional). If set None, the layer
C
chengduo 已提交
1923
            will be named automatically. Default: None
C
chengduoZH 已提交
1924 1925

    Returns:
G
guosheng 已提交
1926
        Variable: The tensor variable storing the convolution and \
C
chengduoZH 已提交
1927 1928
                  non-linearity activation result.

C
refine  
chengduoZH 已提交
1929
    Raises:
1930 1931
        ValueError: If the shapes of input, filter_size, stride, padding and
                    groups mismatch.
C
refine  
chengduoZH 已提交
1932

C
chengduoZH 已提交
1933 1934 1935
    Examples:
        .. code-block:: python

1936 1937
          data = fluid.layers.data(name='data', shape=[3, 32, 32], dtype='float32')
          conv2d = fluid.layers.conv2d(input=data, num_filters=2, filter_size=3, act="relu")
Y
Yu Yang 已提交
1938 1939 1940
    """

    num_channels = input.shape[1]
C
chengduo 已提交
1941
    assert param_attr is not False, "param_attr should not be False here."
1942
    l_type = 'conv2d'
X
xzl 已提交
1943 1944
    if (num_channels == groups and num_filters % num_channels == 0 and
            not use_cudnn):
1945
        l_type = 'depthwise_conv2d'
1946 1947 1948 1949

    helper = LayerHelper(l_type, **locals())
    dtype = helper.input_dtype()

Y
Yu Yang 已提交
1950 1951 1952 1953 1954
    if groups is None:
        num_filter_channels = num_channels
    else:
        if num_channels % groups != 0:
            raise ValueError("num_channels must be divisible by groups.")
M
minqiyang 已提交
1955
        num_filter_channels = num_channels // groups
Y
Yu Yang 已提交
1956

C
chengduoZH 已提交
1957 1958 1959
    filter_size = utils.convert_to_list(filter_size, 2, 'filter_size')
    stride = utils.convert_to_list(stride, 2, 'stride')
    padding = utils.convert_to_list(padding, 2, 'padding')
1960
    dilation = utils.convert_to_list(dilation, 2, 'dilation')
C
chengduoZH 已提交
1961

C
chengduoZH 已提交
1962 1963
    if not isinstance(use_cudnn, bool):
        raise ValueError("use_cudnn should be True or False")
Y
Yu Yang 已提交
1964 1965

    input_shape = input.shape
M
minqiyang 已提交
1966
    filter_shape = [num_filters, int(num_filter_channels)] + filter_size
Y
Yu Yang 已提交
1967 1968

    def _get_default_param_initializer():
C
chengduo 已提交
1969 1970
        filter_elem_num = filter_size[0] * filter_size[1] * num_channels
        std = (2.0 / filter_elem_num)**0.5
Y
Yu Yang 已提交
1971 1972 1973 1974 1975 1976 1977 1978
        return Normal(0.0, std, 0)

    filter_param = helper.create_parameter(
        attr=helper.param_attr,
        shape=filter_shape,
        dtype=dtype,
        default_initializer=_get_default_param_initializer())

X
Xin Pan 已提交
1979
    pre_bias = helper.create_variable_for_type_inference(dtype)
Y
Yu Yang 已提交
1980

1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994
    if use_cudnn:
        helper.create_variable(
            name="kCUDNNFwdAlgoCache",
            persistable=True,
            type=core.VarDesc.VarType.RAW)
        helper.create_variable(
            name="kCUDNNBwdDataAlgoCache",
            persistable=True,
            type=core.VarDesc.VarType.RAW)
        helper.create_variable(
            name="kCUDNNBwdFilterAlgoCache",
            persistable=True,
            type=core.VarDesc.VarType.RAW)

Y
Yu Yang 已提交
1995
    helper.append_op(
1996
        type=l_type,
Y
Yu Yang 已提交
1997 1998 1999 2000 2001
        inputs={
            'Input': input,
            'Filter': filter_param,
        },
        outputs={"Output": pre_bias},
C
chengduoZH 已提交
2002 2003 2004
        attrs={
            'strides': stride,
            'paddings': padding,
2005
            'dilations': dilation,
C
chengduoZH 已提交
2006
            'groups': groups,
2007
            'use_cudnn': use_cudnn,
2008
            'use_mkldnn': False,
2009
            'fuse_relu_before_depthwise_conv': False
C
chengduoZH 已提交
2010
        })
Y
Yu Yang 已提交
2011 2012 2013 2014 2015 2016

    pre_act = helper.append_bias_op(pre_bias, dim_start=1, dim_end=2)

    return helper.append_activation(pre_act)


C
chengduoZH 已提交
2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033
def conv3d(input,
           num_filters,
           filter_size,
           stride=1,
           padding=0,
           dilation=1,
           groups=None,
           param_attr=None,
           bias_attr=None,
           use_cudnn=True,
           act=None,
           name=None):
    """
    **Convlution3D Layer**

    The convolution3D layer calculates the output based on the input, filter
    and strides, paddings, dilations, groups parameters. Input(Input) and
2034 2035 2036 2037 2038 2039
    Output(Output) are in NCDHW format. Where N is batch size C is the number of
    channels, D is the depth of the feature, H is the height of the feature,
    and W is the width of the feature. Convlution3D is similar with Convlution2D
    but adds one dimension(depth). If bias attribution and activation type are
    provided, bias is added to the output of the convolution, and the
    corresponding activation function is applied to the final result.
C
chengduoZH 已提交
2040 2041 2042 2043 2044 2045 2046 2047 2048

    For each input :math:`X`, the equation is:

    .. math::

        Out = \sigma (W \\ast X + b)

    In the above equation:

2049 2050
    * :math:`X`: Input value, a tensor with NCDHW format.
    * :math:`W`: Filter value, a tensor with MCDHW format.
C
chengduoZH 已提交
2051 2052 2053
    * :math:`\\ast`: Convolution operation.
    * :math:`b`: Bias value, a 2-D tensor with shape [M, 1].
    * :math:`\\sigma`: Activation function.
2054
    * :math:`Out`: Output value, the shape of :math:`Out` and :math:`X` may be different.
C
chengduoZH 已提交
2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079

    Example:

        - Input:

          Input shape: :math:`(N, C_{in}, D_{in}, H_{in}, W_{in})`

          Filter shape: :math:`(C_{out}, C_{in}, D_f, H_f, W_f)`

        - Output:
          Output shape: :math:`(N, C_{out}, D_{out}, H_{out}, W_{out})`

        Where

        .. math::

            D_{out}&= \\frac{(D_{in} + 2 * paddings[0] - (dilations[0] * (D_f - 1) + 1))}{strides[0]} + 1 \\\\
            H_{out}&= \\frac{(H_{in} + 2 * paddings[1] - (dilations[1] * (H_f - 1) + 1))}{strides[1]} + 1 \\\\
            W_{out}&= \\frac{(W_{in} + 2 * paddings[2] - (dilations[2] * (W_f - 1) + 1))}{strides[2]} + 1

    Args:
        input (Variable): The input image with [N, C, D, H, W] format.
            num_filters(int): The number of filter. It is as same as the output
            image channel.
        filter_size (int|tuple|None): The filter size. If filter_size is a tuple,
2080
            it must contain three integers, (filter_size_D, filter_size_H, filter_size_W).
C
chengduoZH 已提交
2081 2082
            Otherwise, the filter will be a square.
        stride (int|tuple): The stride size. If stride is a tuple, it must
2083
            contain three integers, (stride_D, stride_H, stride_W). Otherwise, the
C
chengduoZH 已提交
2084 2085
            stride_D = stride_H = stride_W = stride. Default: stride = 1.
        padding (int|tuple): The padding size. If padding is a tuple, it must
2086
            contain three integers, (padding_D, padding_H, padding_W). Otherwise, the
C
chengduoZH 已提交
2087 2088
            padding_D = padding_H = padding_W = padding. Default: padding = 0.
        dilation (int|tuple): The dilation size. If dilation is a tuple, it must
2089
            contain three integers, (dilation_D, dilation_H, dilation_W). Otherwise, the
C
chengduoZH 已提交
2090 2091 2092 2093 2094 2095
            dilation_D = dilation_H = dilation_W = dilation. Default: dilation = 1.
        groups (int): The groups number of the Conv3d Layer. According to grouped
            convolution in Alex Krizhevsky's Deep CNN paper: when group=2,
            the first half of the filters is only connected to the first half
            of the input channels, while the second half of the filters is only
            connected to the second half of the input channels. Default: groups=1
C
chengduo 已提交
2096 2097 2098 2099 2100 2101 2102 2103 2104 2105
        param_attr (ParamAttr|None): The parameter attribute for learnable parameters/weights
            of conv3d. If it is set to None or one attribute of ParamAttr, conv3d
            will create ParamAttr as param_attr. If it is set to None, the parameter
            is initialized with :math:`Normal(0.0, std)`, and the :math:`std` is
            :math:`(\\frac{2.0 }{filter\_elem\_num})^{0.5}`. Default: None.
        bias_attr (ParamAttr|bool|None): The parameter attribute for the bias of conv3d.
            If it is set to False, no bias will be added to the output units.
            If it is set to None or one attribute of ParamAttr, conv3d
            will create ParamAttr as bias_attr. If the Initializer of the bias_attr
            is not set, the bias is initialized zero. Default: None.
C
chengduoZH 已提交
2106 2107
        use_cudnn (bool): Use cudnn kernel or not, it is valid only when the cudnn
            library is installed. Default: True
C
chengduo 已提交
2108 2109
        act (str): Activation type, if it is set to None, activation is not appended.
            Default: None.
C
chengduoZH 已提交
2110
        name (str|None): A name for this layer(optional). If set None, the layer
C
chengduo 已提交
2111
            will be named automatically. Default: None.
C
chengduoZH 已提交
2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123

    Returns:
        Variable: The tensor variable storing the convolution and \
                  non-linearity activation result.

    Raises:
        ValueError: If the shapes of input, filter_size, stride, padding and
                    groups mismatch.

    Examples:
        .. code-block:: python

2124 2125
          data = fluid.layers.data(name='data', shape=[3, 12, 32, 32], dtype='float32')
          conv3d = fluid.layers.conv3d(input=data, num_filters=2, filter_size=3, act="relu")
C
chengduoZH 已提交
2126 2127 2128
    """

    l_type = 'conv3d'
C
chengduo 已提交
2129
    assert param_attr is not False, "param_attr should not be False here."
C
chengduoZH 已提交
2130 2131 2132 2133 2134 2135 2136 2137 2138 2139
    helper = LayerHelper(l_type, **locals())
    dtype = helper.input_dtype()

    num_channels = input.shape[1]

    if groups is None:
        num_filter_channels = num_channels
    else:
        if num_channels % groups != 0:
            raise ValueError("num_channels must be divisible by groups.")
M
minqiyang 已提交
2140
        num_filter_channels = num_channels // groups
C
chengduoZH 已提交
2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153

    filter_size = utils.convert_to_list(filter_size, 3, 'filter_size')
    stride = utils.convert_to_list(stride, 3, 'stride')
    padding = utils.convert_to_list(padding, 3, 'padding')
    dilation = utils.convert_to_list(dilation, 3, 'dilation')

    if not isinstance(use_cudnn, bool):
        raise ValueError("use_cudnn should be True or False")

    input_shape = input.shape
    filter_shape = [num_filters, num_filter_channels] + filter_size

    def _get_default_param_initializer():
C
chengduo 已提交
2154 2155 2156
        filter_elem_num = filter_size[0] * filter_size[1] * filter_size[
            2] * num_channels
        std = (2.0 / filter_elem_num)**0.5
C
chengduoZH 已提交
2157 2158 2159 2160 2161 2162 2163 2164
        return Normal(0.0, std, 0)

    filter_param = helper.create_parameter(
        attr=helper.param_attr,
        shape=filter_shape,
        dtype=dtype,
        default_initializer=_get_default_param_initializer())

X
Xin Pan 已提交
2165
    pre_bias = helper.create_variable_for_type_inference(dtype)
C
chengduoZH 已提交
2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179

    helper.append_op(
        type=l_type,
        inputs={
            'Input': input,
            'Filter': filter_param,
        },
        outputs={"Output": pre_bias},
        attrs={
            'strides': stride,
            'paddings': padding,
            'dilations': dilation,
            'groups': groups,
            'use_cudnn': use_cudnn,
X
Xin Pan 已提交
2180
            'use_mkldnn': False
C
chengduoZH 已提交
2181 2182
        })

2183
    pre_act = helper.append_bias_op(pre_bias, dim_start=1, dim_end=2)
C
chengduoZH 已提交
2184 2185 2186 2187

    return helper.append_activation(pre_act)


J
Jacek Czaja 已提交
2188
def sequence_pool(input, pool_type, is_test=False):
Y
Yu Yang 已提交
2189
    """
Y
yangyaming 已提交
2190 2191 2192
    This function add the operator for sequence pooling.
    It pools features of all time-steps of each instance, and is applied
    on top of the input using pool_type mentioned in the parameters.
L
Luo Tao 已提交
2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203

    It supports four pool_type:

    - average: :math:`Out[i] = \\frac{\sum_i X_i}{N}`
    - sum:     :math:`Out[i] = \sum_jX_{ij}`
    - sqrt:    :math:`Out[i] = \\frac{\sum_jX_{ij}}{\sqrt{len(X_i)}}`
    - max:     :math:`Out[i] = max(X_i)`

    .. code-block:: text

       x is a 1-level LoDTensor:
2204
         x.lod = [[2, 3, 2]]
L
Luo Tao 已提交
2205 2206 2207 2208 2209
         x.data = [1, 3, 2, 4, 6, 5, 1]
         x.dims = [7, 1]

       then output is a Tensor:
         out.dim = [3, 1]
2210
         with condition len(x.lod[-1]) == out.dims[0]
L
Luo Tao 已提交
2211 2212 2213 2214 2215 2216 2217

       for different pool_type:
         average: out.data = [2, 4, 3], where 2=(1+3)/2, 4=(2+4+6)/3, 3=(5+1)/2
         sum    : out.data = [4, 12, 6], where 4=1+3, 12=2+4+6, 6=5+1
         sqrt   : out.data = [2.82, 6.93, 4.24], where 2.82=(1+3)/sqrt(2),
                    6.93=(2+4+6)/sqrt(3), 4.24=(5+1)/sqrt(2)
         max    : out.data = [3, 6, 5], where 3=max(1,3), 6=max(2,4,6), 5=max(5,1)
2218 2219
         last   : out.data = [3, 6, 1], where 3=last(1,3), 6=last(2,4,6), 1=last(5,1)
         first  : out.data = [1, 2, 5], where 1=first(1,3), 2=first(2,4,6), 5=first(5,1)
F
fengjiayi 已提交
2220

L
Luo Tao 已提交
2221 2222
    Args:
        input(variable): The input variable which is a LoDTensor.
Y
yangyaming 已提交
2223
        pool_type (string): The pooling type of sequence_pool.
L
Luo Tao 已提交
2224
            It supports average, sum, sqrt and max.
J
Jacek Czaja 已提交
2225
        is_test(bool, Default False): Used distinguish training from scoring mode.
L
Luo Tao 已提交
2226 2227 2228 2229 2230 2231 2232

    Returns:
        The sequence pooling variable which is a Tensor.

    Examples:

        .. code-block:: python
F
fengjiayi 已提交
2233

Y
yangyaming 已提交
2234
             x = fluid.layers.data(name='x', shape=[7, 1],
L
Luo Tao 已提交
2235 2236 2237 2238 2239
                              dtype='float32', lod_level=1)
             avg_x = fluid.layers.sequence_pool(input=x, pool_type='average')
             sum_x = fluid.layers.sequence_pool(input=x, pool_type='sum')
             sqrt_x = fluid.layers.sequence_pool(input=x, pool_type='sqrt')
             max_x = fluid.layers.sequence_pool(input=x, pool_type='max')
2240 2241
             last_x = fluid.layers.sequence_pool(input=x, pool_type='last')
             first_x = fluid.layers.sequence_pool(input=x, pool_type='first')
Y
Yu Yang 已提交
2242
    """
F
fengjiayi 已提交
2243
    helper = LayerHelper('sequence_pool', **locals())
Y
Yu Yang 已提交
2244
    dtype = helper.input_dtype()
X
Xin Pan 已提交
2245 2246
    pool_out = helper.create_variable_for_type_inference(dtype)
    max_index = helper.create_variable_for_type_inference(dtype)
Y
Yu Yang 已提交
2247 2248 2249 2250 2251 2252

    helper.append_op(
        type="sequence_pool",
        inputs={"X": input},
        outputs={"Out": pool_out,
                 "MaxIndex": max_index},
J
Jacek Czaja 已提交
2253 2254
        attrs={"pooltype": pool_type.upper(),
               "is_test": is_test})
Y
Yu Yang 已提交
2255

Y
yangyaming 已提交
2256 2257 2258 2259 2260
    # when pool_type is max, variable max_index is initialized,
    # so we stop the gradient explicitly here
    if pool_type == 'max':
        max_index.stop_gradient = True

Y
Yu Yang 已提交
2261 2262 2263
    return pool_out


C
add doc  
chengduoZH 已提交
2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282
@templatedoc()
def sequence_concat(input, name=None):
    """
    ${comment}

    Args:
        input(list): List of Variables to be concatenated.
        name(str|None): A name for this layer(optional). If set None, the layer
                       will be named automatically.

    Returns:
        Variable: Output variable of the concatenation.

    Examples:
        .. code-block:: python

           out = fluid.layers.sequence_concat(input=[seq1, seq2, seq3])
    """
    helper = LayerHelper('sequence_concat', **locals())
X
Xin Pan 已提交
2283
    out = helper.create_variable_for_type_inference(dtype=helper.input_dtype())
C
add doc  
chengduoZH 已提交
2284 2285 2286 2287 2288
    helper.append_op(
        type='sequence_concat', inputs={'X': input}, outputs={'Out': [out]})
    return out


F
fengjiayi 已提交
2289
def sequence_first_step(input):
L
Luo Tao 已提交
2290
    """
L
Luo Tao 已提交
2291
    This function gets the first step of sequence.
L
Luo Tao 已提交
2292 2293 2294 2295

    .. code-block:: text

       x is a 1-level LoDTensor:
2296
         x.lod = [[2, 3, 2]]
L
Luo Tao 已提交
2297 2298 2299 2300 2301
         x.data = [1, 3, 2, 4, 6, 5, 1]
         x.dims = [7, 1]

       then output is a Tensor:
         out.dim = [3, 1]
2302
         with condition len(x.lod[-1]) == out.dims[0]
L
Luo Tao 已提交
2303
         out.data = [1, 2, 5], where 1=first(1,3), 2=first(2,4,6), 5=first(5,1)
F
fengjiayi 已提交
2304

L
Luo Tao 已提交
2305 2306 2307 2308 2309 2310 2311 2312 2313
    Args:
        input(variable): The input variable which is a LoDTensor.

    Returns:
        The sequence's first step variable which is a Tensor.

    Examples:

        .. code-block:: python
F
fengjiayi 已提交
2314

Y
yangyaming 已提交
2315
             x = fluid.layers.data(name='x', shape=[7, 1],
L
Luo Tao 已提交
2316 2317 2318
                              dtype='float32', lod_level=1)
             x_first_step = fluid.layers.sequence_first_step(input=x)
    """
2319 2320 2321
    return sequence_pool(input=input, pool_type="first")


F
fengjiayi 已提交
2322
def sequence_last_step(input):
L
Luo Tao 已提交
2323
    """
L
Luo Tao 已提交
2324
    This function gets the last step of sequence.
L
Luo Tao 已提交
2325 2326 2327 2328

    .. code-block:: text

       x is a 1-level LoDTensor:
2329
         x.lod = [[2, 3, 2]]
L
Luo Tao 已提交
2330 2331 2332 2333 2334
         x.data = [1, 3, 2, 4, 6, 5, 1]
         x.dims = [7, 1]

       then output is a Tensor:
         out.dim = [3, 1]
2335
         with condition len(x.lod[-1]) == out.dims[0]
L
Luo Tao 已提交
2336
         out.data = [3, 6, 1], where 3=last(1,3), 6=last(2,4,6), 1=last(5,1)
F
fengjiayi 已提交
2337

L
Luo Tao 已提交
2338 2339 2340 2341 2342 2343 2344 2345 2346
    Args:
        input(variable): The input variable which is a LoDTensor.

    Returns:
        The sequence's last step variable which is a Tensor.

    Examples:

        .. code-block:: python
F
fengjiayi 已提交
2347

Y
yangyaming 已提交
2348
             x = fluid.layers.data(name='x', shape=[7, 1],
L
Luo Tao 已提交
2349 2350 2351
                              dtype='float32', lod_level=1)
             x_last_step = fluid.layers.sequence_last_step(input=x)
    """
2352 2353 2354
    return sequence_pool(input=input, pool_type="last")


Y
Yibing Liu 已提交
2355 2356 2357 2358
def sequence_slice(input, offset, length, name=None):
    """
    **Sequence Slice Layer**

2359
    The layer crops a subsequence from given sequence with given start
Y
Yibing Liu 已提交
2360 2361 2362 2363 2364
    offset and subsequence length.

    It only supports sequence data (LoDTensor with lod_level equal to 1).

    .. code-block:: text
2365

H
haowang101779990 已提交
2366
              - Case:
Y
Yibing Liu 已提交
2367

2368
            Given the input Variable **input**:
2369

2370 2371 2372
                input.data = [[a1, a2], [b1, b2], [c1, c2], [d1, d2], [e1, e2]],
                input.lod = [[3, 2]],
                input.dims = (5, 2),
Y
Yibing Liu 已提交
2373

2374
            with offset.data = [[0], [1]] and length.data = [[2], [1]],
Y
Yibing Liu 已提交
2375

2376
            the output Variable will be
2377

2378 2379 2380
                out.data = [[a1, a2], [b1, b2], [e1, e2]],
                out.lod = [[2, 1]],
                out.dims = (3, 2).
2381

M
minqiyang 已提交
2382
    Note:
H
haowang101779990 已提交
2383
          The first dimension size of **input**, **offset** and **length**
2384
          should be equal. The **offset** should start from 0.
2385

Y
Yibing Liu 已提交
2386
    Args:
2387
        input(Variable): The input Variable which consists of the complete
Y
Yibing Liu 已提交
2388
                         sequences.
Y
Yibing Liu 已提交
2389 2390 2391 2392 2393 2394
        offset(Variable): The offset to slice each sequence.
        length(Variable): The length of each subsequence.
        name(str|None): A name for this layer(optional). If set None, the
                        layer will be named automatically.

    Returns:
Y
Yibing Liu 已提交
2395
        Variable: The output subsequences.
Y
Yibing Liu 已提交
2396 2397 2398 2399 2400 2401 2402 2403 2404 2405

    Examples:

        .. code-block:: python

             import numpy as np
             seqs = fluid.layers.data(name='x', shape=[10, 5],
                              dtype='float32', lod_level=1)
             offset = fluid.layers.assign(input=np.array([[0, 1]]).astype("int32"))
             length = fluid.layers.assign(input=np.array([[2, 1]]).astype("int32"))
2406
             subseqs = fluid.layers.sequence_slice(input=seqs, offset=offset,
Y
Yibing Liu 已提交
2407 2408 2409 2410
                                                   length=length)
    """
    helper = LayerHelper("sequence_slice", **locals())
    dtype = helper.input_dtype()
X
Xin Pan 已提交
2411
    out = helper.create_variable_for_type_inference(dtype)
Y
Yibing Liu 已提交
2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425

    offset.stop_gradient = True
    length.stop_gradient = True

    helper.append_op(
        type="sequence_slice",
        inputs={"X": input,
                "Offset": offset,
                "Length": length},
        outputs={"Out": out})

    return out


F
fengjiayi 已提交
2426
@templatedoc()
Y
Yu Yang 已提交
2427
def pool2d(input,
C
chengduoZH 已提交
2428 2429
           pool_size=-1,
           pool_type="max",
C
chengduoZH 已提交
2430 2431
           pool_stride=1,
           pool_padding=0,
C
caoying03 已提交
2432
           global_pooling=False,
C
chengduoZH 已提交
2433
           use_cudnn=True,
2434
           ceil_mode=False,
2435 2436
           name=None,
           exclusive=True):
Y
Yu Yang 已提交
2437
    """
F
fengjiayi 已提交
2438
    ${comment}
2439 2440

    Args:
2441 2442 2443
        input (Variable): The input tensor of pooling operator. The format of
                          input tensor is NCHW, where N is batch size, C is
                          the number of channels, H is the height of the
F
fengjiayi 已提交
2444
                          feature, and W is the width of the feature.
J
JiabinYang 已提交
2445
        pool_size (int|list|tuple): The pool kernel size. If pool kernel size is a tuple or list,
J
JiabinYang 已提交
2446 2447
            it must contain two integers, (pool_size_Height, pool_size_Width).
            Otherwise, the pool kernel size will be a square of an int.
F
fengjiayi 已提交
2448
        pool_type: ${pooling_type_comment}
J
JiabinYang 已提交
2449 2450 2451 2452 2453 2454
        pool_stride (int|list|tuple): The pool stride size. If pool stride size is a tuple or list,
            it must contain two integers, (pool_stride_Height, pool_stride_Width).
            Otherwise, the pool stride size will be a square of an int.
        pool_padding (int|list|tuple): The pool padding size. If pool padding size is a tuple,
            it must contain two integers, (pool_padding_on_Height, pool_padding_on_Width).
            Otherwise, the pool padding size will be a square of an int.
2455 2456 2457
        global_pooling (bool): ${global_pooling_comment}
        use_cudnn (bool): ${use_cudnn_comment}
        ceil_mode (bool): ${ceil_mode_comment}
2458
        name (str|None): A name for this layer(optional). If set None, the
F
fengjiayi 已提交
2459
                        layer will be named automatically.
2460
        exclusive (bool): Whether to exclude padding points in average pooling
2461
                          mode, default is true
F
fengjiayi 已提交
2462

2463
    Returns:
F
fengjiayi 已提交
2464
        Variable: The pooling result.
F
fengjiayi 已提交
2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476

    Raises:
        ValueError: If 'pool_type' is not "max" nor "avg"
        ValueError: If 'global_pooling' is False and 'pool_size' is -1
        ValueError: If 'use_cudnn' is not a bool value.

    Examples:

        .. code-block:: python

          data = fluid.layers.data(
              name='data', shape=[3, 32, 32], dtype='float32')
D
dengkaipeng 已提交
2477
          pool2d = fluid.layers.pool2d(
2478 2479 2480 2481
                            input=data,
                            pool_size=2,
                            pool_type='max',
                            pool_stride=1,
F
fengjiayi 已提交
2482
                            global_pooling=False)
Y
Yu Yang 已提交
2483 2484 2485 2486 2487
    """
    if pool_type not in ["max", "avg"]:
        raise ValueError(
            "Unknown pool_type: '%s'. It can only be 'max' or 'avg'.",
            str(pool_type))
C
chengduoZH 已提交
2488

C
chengduoZH 已提交
2489 2490 2491 2492 2493
    if global_pooling is False and pool_size == -1:
        raise ValueError(
            "When the global_pooling is False, pool_size must be passed "
            "and be a valid value. Received pool_size: " + str(pool_size))

C
chengduoZH 已提交
2494 2495 2496 2497
    pool_size = utils.convert_to_list(pool_size, 2, 'pool_size')
    pool_padding = utils.convert_to_list(pool_padding, 2, 'pool_padding')
    pool_stride = utils.convert_to_list(pool_stride, 2, 'pool_stride')

C
chengduoZH 已提交
2498 2499
    if not isinstance(use_cudnn, bool):
        raise ValueError("use_cudnn should be True or False")
Y
Yu Yang 已提交
2500

C
Add doc  
chengduoZH 已提交
2501
    l_type = 'pool2d'
2502 2503

    helper = LayerHelper(l_type, **locals())
Y
Yu Yang 已提交
2504
    dtype = helper.input_dtype()
X
Xin Pan 已提交
2505
    pool_out = helper.create_variable_for_type_inference(dtype)
Y
Yu Yang 已提交
2506 2507

    helper.append_op(
2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518
        type=l_type,
        inputs={"X": input},
        outputs={"Out": pool_out},
        attrs={
            "pooling_type": pool_type,
            "ksize": pool_size,
            "global_pooling": global_pooling,
            "strides": pool_stride,
            "paddings": pool_padding,
            "use_cudnn": use_cudnn,
            "ceil_mode": ceil_mode,
2519 2520
            "use_mkldnn": False,
            "exclusive": exclusive,
2521 2522 2523 2524 2525
        })

    return pool_out


D
dengkaipeng 已提交
2526
@templatedoc()
2527 2528 2529 2530 2531 2532 2533 2534
def pool3d(input,
           pool_size=-1,
           pool_type="max",
           pool_stride=1,
           pool_padding=0,
           global_pooling=False,
           use_cudnn=True,
           ceil_mode=False,
2535 2536
           name=None,
           exclusive=True):
2537
    """
2538
    ${comment}
2539 2540

    Args:
D
dengkaipeng 已提交
2541 2542 2543 2544 2545
        input (Variable): The input tensor of pooling operator. The format of
                          input tensor is NCDHW, where N is batch size, C is
                          the number of channels, D is the depth of the feature,
                          H is the height of the feature, and W is the width
                          of the feature.
D
dengkaipeng 已提交
2546 2547 2548 2549 2550
        pool_size (int|list|tuple): The pool kernel size. If pool kernel size 
            is a tuple or list, it must contain three integers, 
            (pool_size_Depth, pool_size_Height, pool_size_Width).
            Otherwise, the pool kernel size will be the cube of an int.
        pool_type (string): ${pooling_type_comment}
2551 2552 2553 2554 2555 2556 2557
        pool_stride (int): stride of the pooling layer.
        pool_padding (int): padding size.
        global_pooling (bool): ${global_pooling_comment}
        use_cudnn (bool): ${use_cudnn_comment}
        ceil_mode (bool): ${ceil_mode_comment}
        name (str): A name for this layer(optional). If set None, the layer
            will be named automatically.
2558
        exclusive (bool): Whether to exclude padding points in average pooling
2559
                          mode, default is true
2560

2561
    Returns:
2562
        Variable: output of pool3d layer.
D
dengkaipeng 已提交
2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575

    Examples:

        .. code-block:: python

          data = fluid.layers.data(
              name='data', shape=[3, 32, 32, 32], dtype='float32')
          pool3d = fluid.layers.pool3d(
                            input=data,
                            pool_size=2,
                            pool_type='max',
                            pool_stride=1,
                            global_pooling=False)
Y
Yu Yang 已提交
2576 2577 2578 2579 2580
    """
    if pool_type not in ["max", "avg"]:
        raise ValueError(
            "Unknown pool_type: '%s'. It can only be 'max' or 'avg'.",
            str(pool_type))
C
chengduoZH 已提交
2581

C
chengduoZH 已提交
2582 2583 2584 2585 2586
    if global_pooling is False and pool_size == -1:
        raise ValueError(
            "When the global_pooling is False, pool_size must be passed "
            "and be a valid value. Received pool_size: " + str(pool_size))

2587 2588 2589
    pool_size = utils.convert_to_list(pool_size, 3, 'pool_size')
    pool_padding = utils.convert_to_list(pool_padding, 3, 'pool_padding')
    pool_stride = utils.convert_to_list(pool_stride, 3, 'pool_stride')
C
chengduoZH 已提交
2590

C
chengduoZH 已提交
2591 2592
    if not isinstance(use_cudnn, bool):
        raise ValueError("use_cudnn should be True or False")
Y
Yu Yang 已提交
2593

2594 2595
    l_type = "pool3d"
    helper = LayerHelper(l_type, **locals())
Y
Yu Yang 已提交
2596
    dtype = helper.input_dtype()
X
Xin Pan 已提交
2597
    pool_out = helper.create_variable_for_type_inference(dtype)
Y
Yu Yang 已提交
2598 2599

    helper.append_op(
2600
        type=l_type,
Y
Yu Yang 已提交
2601 2602 2603 2604 2605 2606 2607
        inputs={"X": input},
        outputs={"Out": pool_out},
        attrs={
            "pooling_type": pool_type,
            "ksize": pool_size,
            "global_pooling": global_pooling,
            "strides": pool_stride,
C
chengduoZH 已提交
2608
            "paddings": pool_padding,
2609
            "use_cudnn": use_cudnn,
2610
            "ceil_mode": ceil_mode,
2611 2612
            "use_mkldnn": False,
            "exclusive": exclusive,
Y
Yu Yang 已提交
2613 2614 2615 2616 2617
        })

    return pool_out


2618 2619 2620 2621 2622 2623 2624
@templatedoc(op_type="pool2d")
def adaptive_pool2d(input,
                    pool_size,
                    pool_type="max",
                    require_index=False,
                    name=None):
    """
D
dengkaipeng 已提交
2625 2626 2627 2628 2629 2630 2631
    **Adaptive Pool2d Operator**
    The adaptive_pool2d operation calculates the output based on the input, pool_size,
    pool_type parameters. Input(X) and output(Out) are in NCHW format, where N is batch
    size, C is the number of channels, H is the height of the feature, and W is
    the width of the feature. Parameters(pool_size) should contain two elements which
    represent height and width, respectively. Also the H and W dimensions of output(Out)
    is same as Parameter(pool_size).
2632

2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645
    For average adaptive pool2d:

    ..  math::

       hstart &= floor(i * H_{in} / H_{out})

       hend &= ceil((i + 1) * H_{in} / H_{out})

       wstart &= floor(j * W_{in} / W_{out})

       wend &= ceil((j + 1) * W_{in} / W_{out})

       Output(i ,j) &= \\frac{sum(Input[hstart:hend, wstart:wend])}{(hend - hstart) * (wend - wstart)}
2646 2647 2648 2649 2650 2651 2652 2653 2654

    Args:
        input (Variable): The input tensor of pooling operator. The format of
                          input tensor is NCHW, where N is batch size, C is
                          the number of channels, H is the height of the
                          feature, and W is the width of the feature.
        pool_size (int|list|tuple): The pool kernel size. If pool kernel size is a tuple or list,
            it must contain two integers, (pool_size_Height, pool_size_Width).
        pool_type: ${pooling_type_comment}
D
dengkaipeng 已提交
2655 2656
        require_index (bool): If true, the index of max pooling point will be returned along
            with outputs. It cannot be set in average pooling type.
2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670
        name (str|None): A name for this layer(optional). If set None, the
                        layer will be named automatically.

    Returns:
        Variable: The pooling result.

    Raises:
        ValueError: 'pool_type' is not 'max' nor 'avg'.
        ValueError: invalid setting 'require_index' true when 'pool_type' is 'avg'.
        ValueError: 'pool_size' should be a list or tuple with length as 2.

    Examples:
        .. code-block:: python

M
minqiyang 已提交
2671
          # suppose input data in shape of [N, C, H, W], `pool_size` is [m, n],
2672
          # output shape is [N, C, m, n], adaptive pool divide H and W dimentions
M
minqiyang 已提交
2673
          # of input data into m * n grids averagely and performs poolings in each
2674 2675
          # grid to get output.
          # adaptive average pool performs calculations as follow:
M
minqiyang 已提交
2676
          #
2677 2678 2679 2680 2681 2682 2683 2684
          #     for i in range(m):
          #         for j in range(n):
          #             hstart = floor(i * H / m)
          #             hend = ceil((i + 1) * H / m)
          #             wstart = floor(i * W / n)
          #             wend = ceil((i + 1) * W / n)
          #             output[:, :, i, j] = avg(input[:, :, hstart: hend, wstart: wend])
          #
2685 2686
          data = fluid.layers.data(
              name='data', shape=[3, 32, 32], dtype='float32')
D
dengkaipeng 已提交
2687
          pool_out = fluid.layers.adaptive_pool2d(
2688 2689
                            input=data,
                            pool_size=[3, 3],
2690
                            pool_type='avg')
2691 2692 2693 2694 2695 2696 2697 2698 2699 2700
    """
    if pool_type not in ["max", "avg"]:
        raise ValueError(
            "Unknown pool_type: '%s'. It can only be 'max' or 'avg'.",
            str(pool_type))

    if pool_type == "avg" and require_index:
        raise ValueError(
            "invalid setting 'require_index' true when 'pool_type' is 'avg'.")

2701
    pool_size = utils.convert_to_list(pool_size, 2, 'pool_size')
2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726

    if pool_type == "max":
        l_type = 'max_pool2d_with_index'
    else:
        l_type = "pool2d"

    helper = LayerHelper(l_type, **locals())
    dtype = helper.input_dtype()
    pool_out = helper.create_variable_for_type_inference(dtype)

    outputs = {"Out": pool_out}
    if pool_type == "max":
        mask = helper.create_variable_for_type_inference(dtype)
        outputs["Mask"] = mask

    helper.append_op(
        type=l_type,
        inputs={"X": input},
        outputs=outputs,
        attrs={
            "pooling_type": pool_type,
            "ksize": pool_size,
            "adaptive": True,
        })

D
dengkaipeng 已提交
2727
    return (pool_out, mask) if require_index else pool_out
2728 2729 2730 2731 2732 2733 2734 2735 2736


@templatedoc(op_type="pool3d")
def adaptive_pool3d(input,
                    pool_size,
                    pool_type="max",
                    require_index=False,
                    name=None):
    """
D
dengkaipeng 已提交
2737 2738 2739 2740 2741 2742 2743
    **Adaptive Pool3d Operator**
    The adaptive_pool3d operation calculates the output based on the input, pool_size,
    pool_type parameters. Input(X) and output(Out) are in NCDHW format, where N is batch
    size, C is the number of channels, D is the depth of the feature, H is the height of
    the feature, and W is the width of the feature. Parameters(pool_size) should contain
    three elements which represent height and width, respectively. Also the D, H and W
    dimensions of output(Out) is same as Parameter(pool_size).
2744

2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761
    For average adaptive pool3d:

    ..  math::

      dstart &= floor(i * D_{in} / D_{out})

      dend &= ceil((i + 1) * D_{in} / D_{out})

      hstart &= floor(j * H_{in} / H_{out})

      hend &= ceil((j + 1) * H_{in} / H_{out})

      wstart &= floor(k * W_{in} / W_{out})

      wend &= ceil((k + 1) * W_{in} / W_{out})

      Output(i ,j, k) &= \\frac{sum(Input[dstart:dend, hstart:hend, wstart:wend])}{(dend - dstart) * (hend - hstart) * (wend - wstart)}
2762 2763 2764

    Args:
        input (Variable): The input tensor of pooling operator. The format of
D
dengkaipeng 已提交
2765 2766 2767
                          input tensor is NCDHW, where N is batch size, C is
                          the number of channels, D is the depth of the feature,
                          H is the height of the feature, and W is the width of the feature.
2768
        pool_size (int|list|tuple): The pool kernel size. If pool kernel size is a tuple or list,
D
dengkaipeng 已提交
2769
            it must contain three integers, (Depth, Height, Width).
2770
        pool_type: ${pooling_type_comment}
D
dengkaipeng 已提交
2771 2772
        require_index (bool): If true, the index of max pooling point will be returned along
            with outputs. It cannot be set in average pooling type.
2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786
        name (str|None): A name for this layer(optional). If set None, the
                        layer will be named automatically.

    Returns:
        Variable: The pooling result.

    Raises:
        ValueError: 'pool_type' is not 'max' nor 'avg'.
        ValueError: invalid setting 'require_index' true when 'pool_type' is 'avg'.
        ValueError: 'pool_size' should be a list or tuple with length as 2.

    Examples:
        .. code-block:: python

2787 2788
          # suppose input data in shape of [N, C, D, H, W], `pool_size` is [l, m, n],
          # output shape is [N, C, l, m, n], adaptive pool divide D, H and W dimentions
M
minqiyang 已提交
2789
          # of input data into l * m * n grids averagely and performs poolings in each
2790 2791
          # grid to get output.
          # adaptive average pool performs calculations as follow:
M
minqiyang 已提交
2792
          #
2793 2794 2795 2796 2797 2798 2799 2800 2801
          #     for i in range(l):
          #         for j in range(m):
          #             for k in range(n):
          #                 dstart = floor(i * D / l)
          #                 dend = ceil((i + 1) * D / l)
          #                 hstart = floor(j * H / m)
          #                 hend = ceil((j + 1) * H / m)
          #                 wstart = floor(k * W / n)
          #                 wend = ceil((k + 1) * W / n)
M
minqiyang 已提交
2802
          #                 output[:, :, i, j, k] =
2803 2804
          #                     avg(input[:, :, dstart:dend, hstart: hend, wstart: wend])
          #
2805 2806
          data = fluid.layers.data(
              name='data', shape=[3, 32, 32], dtype='float32')
D
dengkaipeng 已提交
2807
          pool_out, mask = fluid.layers.adaptive_pool3d(
2808
                            input=data,
D
dengkaipeng 已提交
2809
                            pool_size=[3, 3, 3],
2810
                            pool_type='avg')
2811 2812 2813 2814 2815 2816 2817 2818 2819 2820
    """
    if pool_type not in ["max", "avg"]:
        raise ValueError(
            "Unknown pool_type: '%s'. It can only be 'max' or 'avg'.",
            str(pool_type))

    if pool_type == "avg" and require_index:
        raise ValueError(
            "invalid setting 'require_index' true when 'pool_type' is 'avg'.")

2821
    pool_size = utils.convert_to_list(pool_size, 3, 'pool_size')
2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846

    if pool_type == "max":
        l_type = 'max_pool3d_with_index'
    else:
        l_type = "pool3d"

    helper = LayerHelper(l_type, **locals())
    dtype = helper.input_dtype()
    pool_out = helper.create_variable_for_type_inference(dtype)

    outputs = {"Out": pool_out}
    if pool_type == "max":
        mask = helper.create_variable_for_type_inference(dtype)
        outputs["Mask"] = mask

    helper.append_op(
        type=l_type,
        inputs={"X": input},
        outputs=outputs,
        attrs={
            "pooling_type": pool_type,
            "ksize": pool_size,
            "adaptive": True,
        })

D
dengkaipeng 已提交
2847
    return (pool_out, mask) if require_index else pool_out
2848 2849


Y
Yu Yang 已提交
2850 2851 2852 2853 2854 2855 2856
def batch_norm(input,
               act=None,
               is_test=False,
               momentum=0.9,
               epsilon=1e-05,
               param_attr=None,
               bias_attr=None,
C
caoying03 已提交
2857
               data_layout='NCHW',
Y
Yang Yang 已提交
2858
               in_place=False,
2859 2860
               name=None,
               moving_mean_name=None,
W
wanghaoshuang 已提交
2861
               moving_variance_name=None,
2862
               do_model_average_for_mean_and_var=False,
2863 2864
               fuse_with_relu=False,
               use_global_stats=False):
Y
Yu Yang 已提交
2865
    """
Q
qiaolongfei 已提交
2866 2867 2868 2869
    **Batch Normalization Layer**

    Can be used as a normalizer function for conv2d and fully_connected operations.
    The required data format for this layer is one of the following:
Q
qiaolongfei 已提交
2870

Q
qiaolongfei 已提交
2871
    1. NHWC `[batch, in_height, in_width, in_channels]`
Q
qiaolongfei 已提交
2872

Q
qiaolongfei 已提交
2873 2874
    2. NCHW `[batch, in_channels, in_height, in_width]`

Q
qiaolongfei 已提交
2875 2876 2877
    Refer to `Batch Normalization: Accelerating Deep Network Training by Reducing
    Internal Covariate Shift <https://arxiv.org/pdf/1502.03167.pdf>`_
    for more details.
Q
qiaolongfei 已提交
2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889

    :math:`input` is the input features over a mini-batch.

    ..  math::

        \\mu_{\\beta} &\\gets \\frac{1}{m} \\sum_{i=1}^{m} x_i \\qquad &//\\
        \ mini-batch\ mean \\\\
        \\sigma_{\\beta}^{2} &\\gets \\frac{1}{m} \\sum_{i=1}^{m}(x_i - \\
        \\mu_{\\beta})^2 \\qquad &//\ mini-batch\ variance \\\\
        \\hat{x_i} &\\gets \\frac{x_i - \\mu_\\beta} {\\sqrt{\\
        \\sigma_{\\beta}^{2} + \\epsilon}} \\qquad &//\ normalize \\\\
        y_i &\\gets \\gamma \\hat{x_i} + \\beta \\qquad &//\ scale\ and\ shift
2890

2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903

    When use_global_stats = True, the :math:`\\mu_{\\beta}`
    and :math:`\\sigma_{\\beta}^{2}` are not the statistics of one mini-batch.
    They are global (or running) statistics. (It usually got from the
    pre-trained model.)
    The training and testing (or inference) have the same behavior:

    ..  math::

        \\hat{x_i} &\\gets \\frac{x_i - \\mu_\\beta} {\\sqrt{\\
        \\sigma_{\\beta}^{2} + \\epsilon}}  \\\\
        y_i &\\gets \\gamma \\hat{x_i} + \\beta

2904
    Args:
Q
qiaolongfei 已提交
2905
        input(variable): The input variable which is a LoDTensor.
Q
qiaolongfei 已提交
2906 2907 2908 2909
        act(string, Default None): Activation type, linear|relu|prelu|...
        is_test(bool, Default False): Used for training or training.
        momentum(float, Default 0.9):
        epsilon(float, Default 1e-05):
C
chengduo 已提交
2910 2911 2912 2913 2914 2915 2916 2917
        param_attr(ParamAttr|None): The parameter attribute for Parameter `scale`
             of batch_norm. If it is set to None or one attribute of ParamAttr, batch_norm
             will create ParamAttr as param_attr. If the Initializer of the param_attr
             is not set, the parameter is initialized with Xavier. Default: None.
        bias_attr(ParamAttr|None): The parameter attribute for the bias of batch_norm.
             If it is set to None or one attribute of ParamAttr, batch_norm
             will create ParamAttr as bias_attr. If the Initializer of the bias_attr
             is not set, the bias is initialized zero. Default: None.
Q
qiaolongfei 已提交
2918
        data_layout(string, default NCHW): NCHW|NHWC
2919
        in_place(bool, Default False): Make the input and output of batch norm reuse memory.
Q
qiaolongfei 已提交
2920 2921 2922 2923
        name(string, Default None): A name for this layer(optional). If set None, the layer
            will be named automatically.
        moving_mean_name(string, Default None): The name of moving_mean which store the global Mean.
        moving_variance_name(string, Default None): The name of the moving_variance which store the global Variance.
Q
qiaolongfei 已提交
2924
        do_model_average_for_mean_and_var(bool, Default False): Do model average for mean and variance or not.
2925
        fuse_with_relu (bool): if True, this OP performs relu after batch norm.
2926 2927 2928 2929 2930
        use_global_stats(bool, Default False): Whether to use global mean and
            variance. In inference or test mode, set use_global_stats to true
            or is_test to true, and the behavior is equivalent.
            In train mode, when setting use_global_stats True, the global mean
            and variance are also used during train period.
2931 2932

    Returns:
Q
qiaolongfei 已提交
2933
        Variable: A tensor variable which is the result after applying batch normalization on the input.
Q
qiaolongfei 已提交
2934 2935 2936 2937 2938 2939 2940

    Examples:

        .. code-block:: python

            hidden1 = fluid.layers.fc(input=x, size=200, param_attr='fc1.w')
            hidden2 = fluid.layers.batch_norm(input=hidden1)
Y
Yu Yang 已提交
2941
    """
C
chengduo 已提交
2942
    assert bias_attr is not False, "bias_attr should not be False in batch_norm."
Y
Yu Yang 已提交
2943 2944 2945
    helper = LayerHelper('batch_norm', **locals())
    dtype = helper.input_dtype()

W
Wu Yi 已提交
2946 2947 2948 2949
    # use fp32 for bn parameter
    if dtype == core.VarDesc.VarType.FP16:
        dtype = core.VarDesc.VarType.FP32

Y
Yu Yang 已提交
2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966
    input_shape = input.shape
    if data_layout == 'NCHW':
        channel_num = input_shape[1]
    else:
        if data_layout == 'NHWC':
            channel_num = input_shape[-1]
        else:
            raise ValueError("unsupported data layout:" + data_layout)

    param_shape = [channel_num]

    # create parameter
    scale = helper.create_parameter(
        attr=helper.param_attr,
        shape=param_shape,
        dtype=dtype,
        default_initializer=Constant(1.0))
2967 2968 2969
    # setting stop_gradient=True to reduce computation
    if use_global_stats and helper.param_attr.learning_rate == 0.:
        scale.stop_gradient = True
Y
Yu Yang 已提交
2970 2971

    bias = helper.create_parameter(
2972
        attr=helper.bias_attr, shape=param_shape, dtype=dtype, is_bias=True)
2973 2974
    # setting stop_gradient=True to reduce computation
    if use_global_stats and helper.bias_attr.learning_rate == 0.:
M
minqiyang 已提交
2975
        bias.stop_gradient = True
Y
Yu Yang 已提交
2976

2977 2978
    mean = helper.create_parameter(
        attr=ParamAttr(
W
wanghaoshuang 已提交
2979 2980 2981
            name=moving_mean_name,
            initializer=Constant(0.0),
            trainable=False,
W
wanghaoshuang 已提交
2982
            do_model_average=do_model_average_for_mean_and_var),
Q
QI JUN 已提交
2983
        shape=param_shape,
W
Wu Yi 已提交
2984
        dtype=dtype)
2985 2986 2987 2988 2989 2990
    mean.stop_gradient = True

    variance = helper.create_parameter(
        attr=ParamAttr(
            name=moving_variance_name,
            initializer=Constant(1.0),
W
wanghaoshuang 已提交
2991
            trainable=False,
W
wanghaoshuang 已提交
2992
            do_model_average=do_model_average_for_mean_and_var),
Q
QI JUN 已提交
2993
        shape=param_shape,
W
Wu Yi 已提交
2994
        dtype=dtype)
2995
    variance.stop_gradient = True
Y
Yu Yang 已提交
2996 2997 2998 2999 3000 3001

    # create output
    # mean and mean_out share the same memory
    mean_out = mean
    # variance and variance out share the same memory
    variance_out = variance
X
Xin Pan 已提交
3002 3003 3004 3005
    saved_mean = helper.create_variable_for_type_inference(
        dtype=dtype, stop_gradient=True)
    saved_variance = helper.create_variable_for_type_inference(
        dtype=dtype, stop_gradient=True)
Y
Yu Yang 已提交
3006

X
Xin Pan 已提交
3007 3008
    batch_norm_out = input if in_place else helper.create_variable_for_type_inference(
        dtype)
Y
Yu Yang 已提交
3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025

    helper.append_op(
        type="batch_norm",
        inputs={
            "X": input,
            "Scale": scale,
            "Bias": bias,
            "Mean": mean,
            "Variance": variance
        },
        outputs={
            "Y": batch_norm_out,
            "MeanOut": mean_out,
            "VarianceOut": variance_out,
            "SavedMean": saved_mean,
            "SavedVariance": saved_variance
        },
3026 3027 3028 3029
        attrs={
            "momentum": momentum,
            "epsilon": epsilon,
            "is_test": is_test,
3030
            "data_layout": data_layout,
X
Xin Pan 已提交
3031
            "use_mkldnn": False,
3032 3033
            "fuse_with_relu": fuse_with_relu,
            "use_global_stats": use_global_stats
3034
        })
Y
Yu Yang 已提交
3035 3036 3037 3038

    return helper.append_activation(batch_norm_out)


H
heqiaozhi 已提交
3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157
def data_norm(input,
              act=None,
              epsilon=1e-05,
              param_attr=None,
              data_layout='NCHW',
              in_place=False,
              name=None,
              moving_mean_name=None,
              moving_variance_name=None,
              do_model_average_for_mean_and_var=False):
    """
    **Data Normalization Layer**

    Can be used as a normalizer function for conv2d and fully_connected operations.
    The required data format for this layer is one of the following:

    1. NHWC `[batch, in_height, in_width, in_channels]`

    2. NCHW `[batch, in_channels, in_height, in_width]`

    :math:`input` is the input features over a mini-batch.

    ..  math::

        \\mu_{\\beta} &\\gets \\frac{1}{m} \\sum_{i=1}^{m} x_i \\qquad &//\\
        \ mini-batch\ mean \\\\
        \\sigma_{\\beta}^{2} &\\gets \\frac{1}{m} \\sum_{i=1}^{m}(x_i - \\
        \\mu_{\\beta})^2 \\qquad &//\ mini-batch\ variance \\\\
        \\hat{x_i} &\\gets \\frac{x_i - \\mu_\\beta} {\\sqrt{\\
        \\sigma_{\\beta}^{2} + \\epsilon}} \\qquad &//\ normalize \\\\
        y_i &\\gets \\gamma \\hat{x_i} + \\beta \\qquad &//\ scale\ and\ shift

    Args:
        input(variable): The input variable which is a LoDTensor.
        act(string, Default None): Activation type, linear|relu|prelu|...
        epsilon(float, Default 1e-05):
        param_attr(ParamAttr): The parameter attribute for Parameter `scale`.
        data_layout(string, default NCHW): NCHW|NHWC
        in_place(bool, Default False): Make the input and output of batch norm reuse memory.
        name(string, Default None): A name for this layer(optional). If set None, the layer
            will be named automatically.
        moving_mean_name(string, Default None): The name of moving_mean which store the global Mean.
        moving_variance_name(string, Default None): The name of the moving_variance which store the global Variance.
        do_model_average_for_mean_and_var(bool, Default False): Do model average for mean and variance or not.

    Returns:
        Variable: A tensor variable which is the result after applying data normalization on the input.

    Examples:

        .. code-block:: python

            data = fluid.layers.data(input=x, size=200, param_attr='fc1.w')
            hidden2 = fluid.layers.data_norm(input=hidden1)
    """
    helper = LayerHelper('data_norm', **locals())
    dtype = helper.input_dtype()

    input_shape = input.shape
    if data_layout == 'NCHW':
        channel_num = input_shape[1]
    else:
        if data_layout == 'NHWC':
            channel_num = input_shape[-1]
        else:
            raise ValueError("unsupported data layout:" + data_layout)

    param_shape = [channel_num]

    batch_size_default = 1e4
    batch_sum_default = 0.0
    batch_square_sum_default = 1e4

    if param_attr and isinstance(param_attr, dict):
        batch_size_default = param_attr.get("batch_size", 1e4)
        batch_sum_default = param_attr.get("batch_sum", 0.0)
        batch_square_sum_default = param_attr.get("batch_square", 1e4)

    # create parameter
    batch_size = helper.create_parameter(
        attr=ParamAttr(
            name=name + '.batch_size',
            initializer=Constant(value=float(batch_size_default)),
            trainable=True),
        shape=param_shape,
        dtype=input.dtype)

    batch_sum = helper.create_parameter(
        attr=ParamAttr(
            name=name + '.batch_sum',
            initializer=Constant(value=float(batch_sum_default)),
            trainable=True),
        shape=param_shape,
        dtype=input.dtype)

    batch_square_sum = helper.create_parameter(
        attr=ParamAttr(
            name=name + '.batch_square_sum',
            initializer=Constant(value=float(batch_square_sum_default)),
            trainable=True),
        shape=param_shape,
        dtype=input.dtype)

    means = helper.create_variable(dtype=dtype, stop_gradient=True)
    scales = helper.create_variable(dtype=dtype, stop_gradient=True)

    data_norm_out = input if in_place else helper.create_variable(dtype=dtype)

    helper.append_op(
        type="data_norm",
        inputs={
            "X": input,
            "BatchSize": batch_size,
            "BatchSum": batch_sum,
            "BatchSquareSum": batch_square_sum
        },
        outputs={"Y": data_norm_out,
                 "Means": means,
                 "Scales": scales},
H
heqiaozhi 已提交
3158
        attrs={"epsilon": epsilon})
H
heqiaozhi 已提交
3159 3160 3161 3162

    return helper.append_activation(data_norm_out)


Y
yuyang18 已提交
3163
@templatedoc()
G
guosheng 已提交
3164 3165 3166 3167 3168 3169 3170 3171 3172 3173
def layer_norm(input,
               scale=True,
               shift=True,
               begin_norm_axis=1,
               epsilon=1e-05,
               param_attr=None,
               bias_attr=None,
               act=None,
               name=None):
    """
Y
yuyang18 已提交
3174
    ${comment}
G
guosheng 已提交
3175 3176 3177

    The formula is as follows:

Y
yuyang18 已提交
3178
    ..  math::
G
guosheng 已提交
3179 3180 3181 3182 3183 3184 3185

        \\mu & = \\frac{1}{H}\\sum_{i=1}^{H} a_i

        \\sigma & = \\sqrt{\\frac{1}{H}\sum_{i=1}^{H}(a_i - \\mu)^2}

        h & = f(\\frac{g}{\\sigma}(a - \\mu) + b)

Y
yuyang18 已提交
3186 3187 3188 3189 3190 3191 3192 3193
    * :math:`a`: the vector representation of the summed inputs to the neurons
    in that layer.

    * :math:`H`: the number of hidden units in a layers

    * :math:`g`: the trainable scale parameter.

    * :math:`b`: the trainable bias parameter.
Y
yuyang18 已提交
3194

G
guosheng 已提交
3195 3196
    Args:
        input(Variable): The input tensor variable.
3197
        scale(bool): Whether to learn the adaptive gain :math:`g` after
S
sneaxiy 已提交
3198
            normalization. Default True.
3199
        shift(bool): Whether to learn the adaptive bias :math:`b` after
S
sneaxiy 已提交
3200 3201
            normalization. Default True.
        begin_norm_axis(int): The normalization will be performed along
G
guosheng 已提交
3202
            dimensions from :attr:`begin_norm_axis` to :attr:`rank(input)`.
S
sneaxiy 已提交
3203
            Default 1.
3204
        epsilon(float): The small value added to the variance to prevent
S
sneaxiy 已提交
3205
            division by zero. Default 1e-05.
G
guosheng 已提交
3206
        param_attr(ParamAttr|None): The parameter attribute for the learnable
S
sneaxiy 已提交
3207 3208
            gain :math:`g`. If :attr:`scale` is False, :attr:`param_attr` is
            omitted. If :attr:`scale` is True and :attr:`param_attr` is None,
3209 3210
            a default :code:`ParamAttr` would be added as scale. The
            :attr:`param_attr` is initialized as 1 if it is added. Default None.
G
guosheng 已提交
3211
        bias_attr(ParamAttr|None): The parameter attribute for the learnable
S
sneaxiy 已提交
3212 3213
            bias :math:`b`. If :attr:`shift` is False, :attr:`bias_attr` is
            omitted. If :attr:`shift` is True and :attr:`param_attr` is None,
3214
            a default :code:`ParamAttr` would be added as bias. The
S
sneaxiy 已提交
3215
            :attr:`bias_attr` is initialized as 0 if it is added. Default None.
G
guosheng 已提交
3216
        act(str): Activation to be applied to the output of layer normalizaiton.
S
sneaxiy 已提交
3217 3218 3219
                  Default None.
        name(str): The name of this layer. It is optional. Default None, and a
                   unique name would be generated automatically.
G
guosheng 已提交
3220 3221

    Returns:
Y
yuyang18 已提交
3222
        ${y_comment}
G
guosheng 已提交
3223 3224 3225

    Examples:

Y
yuyang18 已提交
3226 3227 3228
        >>> data = fluid.layers.data(name='data', shape=[3, 32, 32],
        >>>                          dtype='float32')
        >>> x = fluid.layers.layer_norm(input=data, begin_norm_axis=1)
G
guosheng 已提交
3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243
    """
    helper = LayerHelper('layer_norm', **locals())
    dtype = helper.input_dtype()

    # create intput and parameters
    inputs = {'X': input}
    input_shape = input.shape
    param_shape = [reduce(lambda x, y: x * y, input_shape[begin_norm_axis:])]
    if scale:
        scale = helper.create_parameter(
            attr=helper.param_attr,
            shape=param_shape,
            dtype=dtype,
            default_initializer=Constant(1.0))
        inputs['Scale'] = scale
G
guosheng 已提交
3244
    if shift:
G
guosheng 已提交
3245 3246 3247 3248 3249 3250
        assert bias_attr is not False
        bias = helper.create_parameter(
            attr=helper.bias_attr, shape=param_shape, dtype=dtype, is_bias=True)
        inputs['Bias'] = bias

    # create output
X
Xin Pan 已提交
3251 3252 3253 3254 3255
    mean_out = helper.create_variable_for_type_inference(
        dtype=dtype, stop_gradient=True)
    variance_out = helper.create_variable_for_type_inference(
        dtype=dtype, stop_gradient=True)
    layer_norm_out = helper.create_variable_for_type_inference(dtype)
G
guosheng 已提交
3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270

    helper.append_op(
        type="layer_norm",
        inputs=inputs,
        outputs={
            "Y": layer_norm_out,
            "Mean": mean_out,
            "Variance": variance_out,
        },
        attrs={"epsilon": epsilon,
               "begin_norm_axis": begin_norm_axis})

    return helper.append_activation(layer_norm_out)


D
Dun 已提交
3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282
@templatedoc()
def group_norm(input,
               groups,
               epsilon=1e-05,
               param_attr=None,
               bias_attr=None,
               act=None,
               data_layout='NCHW',
               name=None):
    """
    **Group Normalization Layer**

H
haowang101779990 已提交
3283
    Refer to `Group Normalization <https://arxiv.org/abs/1803.08494>`_ .
D
Dun 已提交
3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330

    Args:
        input(Variable): The input tensor variable.
        groups(int): The number of groups that divided from channels.
        epsilon(float): The small value added to the variance to prevent
            division by zero.
        param_attr(ParamAttr|None): The parameter attribute for the learnable
            scale :math:`g`. If it is set to False, no scale will be added to the output units.
            If it is set to None, the bias is initialized one. Default: None.
        bias_attr(ParamAttr|None): The parameter attribute for the learnable
            bias :math:`b`. If it is set to False, no bias will be added to the output units.
            If it is set to None, the bias is initialized zero. Default: None.
        act(str): Activation to be applied to the output of group normalizaiton.
        data_layout(string|NCHW): Only NCHW is supported.
        name (str): The name of this layer. It is optional.

    Returns:
        Variable: A tensor variable which is the result after applying group normalization on the input.

    Examples:

        >>> data = fluid.layers.data(name='data', shape=[8, 32, 32],
        >>>                          dtype='float32')
        >>> x = fluid.layers.group_norm(input=data, groups=4)
    """
    helper = LayerHelper('group_norm', **locals())
    dtype = helper.input_dtype()

    # create intput and parameters
    inputs = {'X': input}
    input_shape = input.shape
    if data_layout != 'NCHW':
        raise ValueError("unsupported data layout:" + data_layout)
    param_shape = [input_shape[1]]
    if param_attr:
        scale = helper.create_parameter(
            attr=helper.param_attr,
            shape=param_shape,
            dtype=dtype,
            default_initializer=Constant(1.0))
        inputs['Scale'] = scale
    if bias_attr:
        bias = helper.create_parameter(
            attr=helper.bias_attr, shape=param_shape, dtype=dtype, is_bias=True)
        inputs['Bias'] = bias

    # create output
H
heqiaozhi 已提交
3331 3332
    mean_out = helper.create_variable(dtype=dtype, stop_gradient=True)
    variance_out = helper.create_variable(dtype=dtype, stop_gradient=True)
D
Dun 已提交
3333
    group_norm_out = helper.create_variable(dtype=dtype)
D
Dun 已提交
3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348

    helper.append_op(
        type="group_norm",
        inputs=inputs,
        outputs={
            "Y": group_norm_out,
            "Mean": mean_out,
            "Variance": variance_out,
        },
        attrs={"epsilon": epsilon,
               "groups": groups})

    return helper.append_activation(group_norm_out)


Y
Yu Yang 已提交
3349 3350 3351 3352
def conv2d_transpose(input,
                     num_filters,
                     output_size=None,
                     filter_size=None,
C
chengduoZH 已提交
3353 3354 3355
                     padding=0,
                     stride=1,
                     dilation=1,
3356
                     groups=None,
C
caoying03 已提交
3357
                     param_attr=None,
3358
                     bias_attr=None,
C
chengduoZH 已提交
3359
                     use_cudnn=True,
3360
                     act=None,
C
caoying03 已提交
3361
                     name=None):
Y
Yu Yang 已提交
3362
    """
3363 3364 3365 3366 3367 3368 3369 3370
    **Convlution2D transpose layer**

    The convolution2D transpose layer calculates the output based on the input,
    filter, and dilations, strides, paddings. Input(Input) and output(Output)
    are in NCHW format. Where N is batch size, C is the number of channels,
    H is the height of the feature, and W is the width of the feature.
    Parameters(dilations, strides, paddings) are two elements. These two elements
    represent height and width, respectively. The details of convolution transpose
3371 3372
    layer, please refer to the following explanation and references
    `therein <http://www.matthewzeiler.com/wp-content/uploads/2017/07/cvpr2010.pdf>`_.
3373 3374 3375
    If bias attribution and activation type are provided, bias is added to
    the output of the convolution, and the corresponding activation function
    is applied to the final result.
3376 3377 3378 3379 3380

    For each input :math:`X`, the equation is:

    .. math::

3381
        Out = \sigma (W \\ast X + b)
3382

3383
    Where:
3384 3385 3386

    * :math:`X`: Input value, a tensor with NCHW format.
    * :math:`W`: Filter value, a tensor with MCHW format.
3387 3388 3389 3390
    * :math:`\\ast`: Convolution operation.
    * :math:`b`: Bias value, a 2-D tensor with shape [M, 1].
    * :math:`\\sigma`: Activation function.
    * :math:`Out`: Output value, the shape of :math:`Out` and :math:`X` may be different.
Y
Yu Yang 已提交
3391

3392 3393 3394 3395
    Example:

        - Input:

3396
          Input shape: :math:`(N, C_{in}, H_{in}, W_{in})`
3397

3398
          Filter shape: :math:`(C_{in}, C_{out}, H_f, W_f)`
3399 3400 3401

        - Output:

3402
          Output shape: :math:`(N, C_{out}, H_{out}, W_{out})`
3403 3404

        Where
Y
Yu Yang 已提交
3405

3406 3407
        .. math::

3408 3409
           H^\prime_{out} &= (H_{in} - 1) * strides[0] - 2 * paddings[0] + dilations[0] * (H_f - 1) + 1 \\\\
           W^\prime_{out} &= (W_{in} - 1) * strides[1] - 2 * paddings[1] + dilations[1] * (W_f - 1) + 1 \\\\
H
haowang101779990 已提交
3410 3411
           H_{out} &\in [ H^\prime_{out}, H^\prime_{out} + strides[0] ) \\\\
           W_{out} &\in [ W^\prime_{out}, W^\prime_{out} + strides[1] )
Y
Yu Yang 已提交
3412 3413

    Args:
3414 3415 3416 3417
        input(Variable): The input image with [N, C, H, W] format.
        num_filters(int): The number of the filter. It is as same as the output
            image channel.
        output_size(int|tuple|None): The output image size. If output size is a
3418 3419 3420 3421
            tuple, it must contain two integers, (image_H, image_W). None if use
            filter_size, padding, and stride to calculate output_size.
            if output_size and filter_size are specified at the same time, They
            should follow the formula above.
3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439
        filter_size(int|tuple|None): The filter size. If filter_size is a tuple,
            it must contain two integers, (filter_size_H, filter_size_W).
            Otherwise, the filter will be a square. None if use output size to
            calculate filter_size.
        padding(int|tuple): The padding size. If padding is a tuple, it must
            contain two integers, (padding_H, padding_W). Otherwise, the
            padding_H = padding_W = padding. Default: padding = 0.
        stride(int|tuple): The stride size. If stride is a tuple, it must
            contain two integers, (stride_H, stride_W). Otherwise, the
            stride_H = stride_W = stride. Default: stride = 1.
        dilation(int|tuple): The dilation size. If dilation is a tuple, it must
            contain two integers, (dilation_H, dilation_W). Otherwise, the
            dilation_H = dilation_W = dilation. Default: dilation = 1.
        groups(int): The groups number of the Conv2d transpose layer. Inspired by
            grouped convolution in Alex Krizhevsky's Deep CNN paper, in which
            when group=2, the first half of the filters is only connected to the
            first half of the input channels, while the second half of the
            filters is only connected to the second half of the input channels.
C
chengduo 已提交
3440 3441 3442 3443 3444 3445 3446 3447 3448 3449
            Default: groups = 1.
        param_attr (ParamAttr|None): The parameter attribute for learnable parameters/weights
            of conv2d_transpose. If it is set to None or one attribute of ParamAttr, conv2d_transpose
            will create ParamAttr as param_attr. If the Initializer of the param_attr
            is not set, the parameter is initialized with Xavier. Default: None.
        bias_attr (ParamAttr|bool|None): The parameter attribute for the bias of conv2d_transpose.
            If it is set to False, no bias will be added to the output units.
            If it is set to None or one attribute of ParamAttr, conv2d_transpose
            will create ParamAttr as bias_attr. If the Initializer of the bias_attr
            is not set, the bias is initialized zero. Default: None.
3450
        use_cudnn(bool): Use cudnn kernel or not, it is valid only when the cudnn
C
chengduo 已提交
3451 3452 3453
            library is installed. Default: True.
        act (str): Activation type, if it is set to None, activation is not appended.
            Default: None.
3454
        name(str|None): A name for this layer(optional). If set None, the layer
C
chengduo 已提交
3455
            will be named automatically. Default: True.
Y
Yu Yang 已提交
3456 3457

    Returns:
3458
        Variable: The tensor variable storing the convolution transpose result.
3459 3460

    Raises:
3461 3462
        ValueError: If the shapes of input, filter_size, stride, padding and
                    groups mismatch.
3463 3464 3465 3466

    Examples:
       .. code-block:: python

3467 3468
          data = fluid.layers.data(name='data', shape=[3, 32, 32], dtype='float32')
          conv2d_transpose = fluid.layers.conv2d_transpose(input=data, num_filters=2, filter_size=3)
Y
Yu Yang 已提交
3469
    """
C
chengduo 已提交
3470
    assert param_attr is not False, "param_attr should not be False in conv2d_transpose."
3471 3472 3473 3474 3475 3476 3477 3478
    input_channel = input.shape[1]

    op_type = 'conv2d_transpose'
    if (input_channel == groups and num_filters == input_channel and
            not use_cudnn):
        op_type = 'depthwise_conv2d_transpose'

    helper = LayerHelper(op_type, **locals())
Y
Yu Yang 已提交
3479 3480 3481
    if not isinstance(input, Variable):
        raise TypeError("Input of conv2d_transpose must be Variable")

C
chengduoZH 已提交
3482 3483 3484
    padding = utils.convert_to_list(padding, 2, 'padding')
    stride = utils.convert_to_list(stride, 2, 'stride')
    dilation = utils.convert_to_list(dilation, 2, 'dilation')
G
guosheng 已提交
3485

C
chengduoZH 已提交
3486 3487
    if not isinstance(use_cudnn, bool):
        raise ValueError("use_cudnn should be True or False")
G
guosheng 已提交
3488

Y
Yu Yang 已提交
3489 3490 3491 3492 3493
    if filter_size is None:
        if output_size is None:
            raise ValueError("output_size must be set when filter_size is None")
        if isinstance(output_size, int):
            output_size = [output_size, output_size]
G
guosheng 已提交
3494

Y
Yu Yang 已提交
3495 3496
        h_in = input.shape[2]
        w_in = input.shape[3]
G
guosheng 已提交
3497

C
chengduoZH 已提交
3498
        filter_size_h = (output_size[0] - (h_in - 1) * stride[0] + 2 *
M
minqiyang 已提交
3499
                         padding[0] - 1) // dilation[0] + 1
C
chengduoZH 已提交
3500
        filter_size_w = (output_size[1] - (w_in - 1) * stride[1] + 2 *
M
minqiyang 已提交
3501
                         padding[1] - 1) // dilation[1] + 1
Y
Yu Yang 已提交
3502
        filter_size = [filter_size_h, filter_size_w]
C
chengduoZH 已提交
3503 3504 3505
    else:
        filter_size = utils.convert_to_list(filter_size, 2,
                                            'conv2d_transpose.filter_size')
C
chengduo 已提交
3506

3507 3508 3509 3510 3511 3512 3513
    if output_size is None:
        output_size = []
    elif isinstance(output_size, list) or isinstance(output_size, int):
        output_size = utils.convert_to_list(output_size, 2, 'output_size')
    else:
        raise ValueError("output_size should be list or int")
    padding = utils.convert_to_list(padding, 2, 'padding')
3514
    groups = 1 if groups is None else groups
M
minqiyang 已提交
3515
    filter_shape = [input_channel, num_filters // groups] + filter_size
C
chengduo 已提交
3516

Y
Yu Yang 已提交
3517 3518 3519
    img_filter = helper.create_parameter(
        dtype=input.dtype, shape=filter_shape, attr=helper.param_attr)

X
Xin Pan 已提交
3520
    pre_bias = helper.create_variable_for_type_inference(dtype=input.dtype)
Y
Yu Yang 已提交
3521
    helper.append_op(
3522
        type=op_type,
Y
Yu Yang 已提交
3523 3524
        inputs={'Input': [input],
                'Filter': [img_filter]},
3525
        outputs={'Output': pre_bias},
C
chengduoZH 已提交
3526
        attrs={
3527
            'output_size': output_size,
3528 3529 3530 3531 3532
            'strides': stride,
            'paddings': padding,
            'dilations': dilation,
            'groups': groups,
            'use_cudnn': use_cudnn
Y
Yu Yang 已提交
3533 3534
        })

3535 3536 3537
    pre_act = helper.append_bias_op(pre_bias, dim_start=1, dim_end=2)
    out = helper.append_activation(pre_act)
    return out
Y
Yu Yang 已提交
3538 3539


3540
def conv3d_transpose(input,
Y
Yu Yang 已提交
3541 3542 3543
                     num_filters,
                     output_size=None,
                     filter_size=None,
C
chengduoZH 已提交
3544 3545 3546
                     padding=0,
                     stride=1,
                     dilation=1,
3547
                     groups=None,
C
caoying03 已提交
3548
                     param_attr=None,
3549
                     bias_attr=None,
C
chengduoZH 已提交
3550
                     use_cudnn=True,
3551
                     act=None,
C
caoying03 已提交
3552
                     name=None):
Y
Yu Yang 已提交
3553
    """
3554
    **Convlution3D transpose layer**
3555

3556
    The convolution3D transpose layer calculates the output based on the input,
3557
    filter, and dilations, strides, paddings. Input(Input) and output(Output)
3558 3559 3560 3561 3562 3563
    are in NCDHW format. Where N is batch size, C is the number of channels,
    D is the depth of the feature, H is the height of the feature, and W
    is the width of the feature. Parameters(dilations, strides, paddings) are
    two elements. These two elements represent height and width, respectively.
    The details of convolution transpose layer, please refer to the following
    explanation and references `therein <http://www.matthewzeiler.com/wp-content/uploads/2017/07/cvpr2010.pdf>`_.
3564 3565 3566
    If bias attribution and activation type are provided, bias is added to
    the output of the convolution, and the corresponding activation function
    is applied to the final result.
3567 3568 3569 3570 3571

    For each input :math:`X`, the equation is:

    .. math::

3572
        Out = \sigma (W \\ast X + b)
3573 3574 3575

    In the above equation:

3576 3577
    * :math:`X`: Input value, a tensor with NCDHW format.
    * :math:`W`: Filter value, a tensor with MCDHW format.
3578 3579 3580 3581
    * :math:`\\ast`: Convolution operation.
    * :math:`b`: Bias value, a 2-D tensor with shape [M, 1].
    * :math:`\\sigma`: Activation function.
    * :math:`Out`: Output value, the shape of :math:`Out` and :math:`X` may be different.
Y
Yu Yang 已提交
3582

3583 3584 3585 3586
    Example:

        - Input:

3587
          Input shape: :math:`(N, C_{in}, D_{in}, H_{in}, W_{in})`
3588

3589
          Filter shape: :math:`(C_{in}, C_{out}, D_f, H_f, W_f)`
3590 3591 3592

        - Output:

3593
          Output shape: :math:`(N, C_{out}, D_{out}, H_{out}, W_{out})`
3594 3595

        Where
Y
Yu Yang 已提交
3596

3597 3598
        .. math::

3599 3600 3601
           D_{out} &= (D_{in} - 1) * strides[0] - 2 * paddings[0] + dilations[0] * (D_f - 1) + 1 \\\\
           H_{out} &= (H_{in} - 1) * strides[1] - 2 * paddings[1] + dilations[1] * (H_f - 1) + 1 \\\\
           W_{out} &= (W_{in} - 1) * strides[2] - 2 * paddings[2] + dilations[2] * (W_f - 1) + 1
Y
Yu Yang 已提交
3602 3603

    Args:
3604
        input(Variable): The input image with [N, C, D, H, W] format.
3605 3606 3607
        num_filters(int): The number of the filter. It is as same as the output
            image channel.
        output_size(int|tuple|None): The output image size. If output size is a
3608
            tuple, it must contain three integers, (image_D, image_H, image_W). This
3609 3610
            parameter only works when filter_size is None.
        filter_size(int|tuple|None): The filter size. If filter_size is a tuple,
3611
            it must contain three integers, (filter_size_D, filter_size_H, filter_size_W).
3612 3613 3614
            Otherwise, the filter will be a square. None if use output size to
            calculate filter_size.
        padding(int|tuple): The padding size. If padding is a tuple, it must
3615 3616
            contain three integers, (padding_D, padding_H, padding_W). Otherwise, the
            padding_D = padding_H = padding_W = padding. Default: padding = 0.
3617
        stride(int|tuple): The stride size. If stride is a tuple, it must
3618 3619
            contain three integers, (stride_D, stride_H, stride_W). Otherwise, the
            stride_D = stride_H = stride_W = stride. Default: stride = 1.
3620
        dilation(int|tuple): The dilation size. If dilation is a tuple, it must
3621 3622 3623
            contain three integers, (dilation_D, dilation_H, dilation_W). Otherwise, the
            dilation_D = dilation_H = dilation_W = dilation. Default: dilation = 1.
        groups(int): The groups number of the Conv3d transpose layer. Inspired by
3624 3625 3626 3627 3628
            grouped convolution in Alex Krizhevsky's Deep CNN paper, in which
            when group=2, the first half of the filters is only connected to the
            first half of the input channels, while the second half of the
            filters is only connected to the second half of the input channels.
            Default: groups=1
C
chengduo 已提交
3629 3630 3631 3632 3633 3634 3635 3636 3637
        param_attr (ParamAttr|None): The parameter attribute for learnable parameters/weights
            of conv3d_transpose. If it is set to None or one attribute of ParamAttr, conv3d_transpose
            will create ParamAttr as param_attr. If the Initializer of the param_attr
            is not set, the parameter is initialized with Xavier. Default: None.
        bias_attr (ParamAttr|bool|None): The parameter attribute for the bias of conv3d_transpose.
            If it is set to False, no bias will be added to the output units.
            If it is set to None or one attribute of ParamAttr, conv3d_transpose
            will create ParamAttr as bias_attr. If the Initializer of the bias_attr
            is not set, the bias is initialized zero. Default: None.
3638 3639
        use_cudnn(bool): Use cudnn kernel or not, it is valid only when the cudnn
            library is installed. Default: True
C
chengduo 已提交
3640 3641
        act (str): Activation type, if it is set to None, activation is not appended.
            Default: None.
3642 3643
        name(str|None): A name for this layer(optional). If set None, the layer
            will be named automatically.
Y
Yu Yang 已提交
3644 3645

    Returns:
3646
        Variable: The tensor variable storing the convolution transpose result.
3647 3648

    Raises:
3649 3650
        ValueError: If the shapes of input, filter_size, stride, padding and
                    groups mismatch.
3651 3652 3653 3654

    Examples:
       .. code-block:: python

3655 3656
          data = fluid.layers.data(name='data', shape=[3, 12, 32, 32], dtype='float32')
          conv3d_transpose = fluid.layers.conv3d_transpose(input=data, num_filters=2, filter_size=3)
Y
Yu Yang 已提交
3657
    """
C
chengduo 已提交
3658
    assert param_attr is not False, "param_attr should not be False in conv3d_transpose."
3659 3660
    l_type = "conv3d_transpose"
    helper = LayerHelper(l_type, **locals())
Y
Yu Yang 已提交
3661
    if not isinstance(input, Variable):
3662
        raise TypeError("Input of conv3d_transpose must be Variable")
Y
Yu Yang 已提交
3663 3664
    input_channel = input.shape[1]

3665 3666 3667
    padding = utils.convert_to_list(padding, 3, 'padding')
    stride = utils.convert_to_list(stride, 3, 'stride')
    dilation = utils.convert_to_list(dilation, 3, 'dilation')
C
chengduoZH 已提交
3668

C
chengduoZH 已提交
3669 3670 3671
    if not isinstance(use_cudnn, bool):
        raise ValueError("use_cudnn should be True or False")

Y
Yu Yang 已提交
3672 3673 3674 3675 3676 3677
    if filter_size is None:
        if output_size is None:
            raise ValueError("output_size must be set when filter_size is None")
        if isinstance(output_size, int):
            output_size = [output_size, output_size]

3678 3679 3680
        d_in = input.shape[2]
        h_in = input.shape[3]
        w_in = input.shape[4]
C
chengduoZH 已提交
3681

3682
        filter_size_d = (output_size[0] - (d_in - 1) * stride[0] + 2 *
M
minqiyang 已提交
3683
                         padding[0] - 1) // dilation[0] + 1
3684
        filter_size_h = (output_size[1] - (h_in - 1) * stride[1] + 2 *
M
minqiyang 已提交
3685
                         padding[1] - 1) // dilation[1] + 1
3686
        filter_size_w = (output_size[2] - (w_in - 1) * stride[2] + 2 *
M
minqiyang 已提交
3687
                         padding[2] - 1) // dilation[2] + 1
3688
        filter_size = [filter_size_d, filter_size_h, filter_size_w]
C
chengduoZH 已提交
3689
    else:
3690 3691
        filter_size = utils.convert_to_list(filter_size, 3,
                                            'conv3d_transpose.filter_size')
Y
Yu Yang 已提交
3692

3693
    groups = 1 if groups is None else groups
M
minqiyang 已提交
3694
    filter_shape = [input_channel, num_filters // groups] + filter_size
Y
Yu Yang 已提交
3695 3696 3697
    img_filter = helper.create_parameter(
        dtype=input.dtype, shape=filter_shape, attr=helper.param_attr)

X
Xin Pan 已提交
3698
    pre_bias = helper.create_variable_for_type_inference(dtype=input.dtype)
Y
Yu Yang 已提交
3699
    helper.append_op(
3700
        type=l_type,
Y
Yu Yang 已提交
3701 3702
        inputs={'Input': [input],
                'Filter': [img_filter]},
3703
        outputs={'Output': pre_bias},
C
chengduoZH 已提交
3704 3705 3706 3707
        attrs={
            'strides': stride,
            'paddings': padding,
            'dilations': dilation,
3708
            'groups': groups,
C
chengduoZH 已提交
3709 3710
            'use_cudnn': use_cudnn
        })
Y
Yu Yang 已提交
3711

3712 3713
    pre_act = helper.append_bias_op(pre_bias, dim_start=1, dim_end=2)
    out = helper.append_activation(pre_act)
Y
Yu Yang 已提交
3714
    return out
Y
yangyaming 已提交
3715 3716


Y
yangyaming 已提交
3717
def sequence_expand(x, y, ref_level=-1, name=None):
3718
    """Sequence Expand Layer. This layer will expand the input variable **x**
Y
yangyaming 已提交
3719 3720 3721 3722
    according to specified level lod of **y**. Please note that lod level of
    **x** is at most 1 and rank of **x** is at least 2. When rank of **x**
    is greater than 2, then it would be viewed as a 2-D tensor.
    Following examples will explain how sequence_expand works:
3723 3724 3725 3726 3727

    .. code-block:: text

        * Case 1
            x is a LoDTensor:
3728
                x.lod  = [[2,        2]]
Y
yangyaming 已提交
3729
                x.data = [[a], [b], [c], [d]]
3730 3731 3732
                x.dims = [4, 1]

            y is a LoDTensor:
3733 3734
                y.lod = [[2,    2],
                         [3, 3, 1, 1]]
3735

Y
yangyaming 已提交
3736
            ref_level: 0
3737

Y
yangyaming 已提交
3738
            then output is a 1-level LoDTensor:
3739
                out.lod =  [[2,        2,        2,        2]]
Y
yangyaming 已提交
3740
                out.data = [[a], [b], [a], [b], [c], [d], [c], [d]]
3741 3742 3743 3744
                out.dims = [8, 1]

        * Case 2
            x is a Tensor:
Y
yangyaming 已提交
3745
                x.data = [[a], [b], [c]]
3746 3747 3748
                x.dims = [3, 1]

            y is a LoDTensor:
3749
                y.lod = [[2, 0, 3]]
3750

Y
yangyaming 已提交
3751
            ref_level: -1
3752

Y
yangyaming 已提交
3753 3754 3755
            then output is a Tensor:
                out.data = [[a], [a], [c], [c], [c]]
                out.dims = [5, 1]
3756 3757 3758
    Args:
        x (Variable): The input variable which is a Tensor or LoDTensor.
        y (Variable): The input variable which is a LoDTensor.
Y
yangyaming 已提交
3759 3760
        ref_level (int): Lod level of `y` to be referred by `x`. If set to -1,
                         refer the last level of lod.
C
caoying03 已提交
3761
        name(str|None): A name for this layer(optional). If set None, the layer
Y
yangyaming 已提交
3762
                        will be named automatically.
3763 3764 3765 3766 3767 3768 3769 3770 3771 3772

    Returns:
        Variable: The expanded variable which is a LoDTensor.

    Examples:
        .. code-block:: python

            x = fluid.layers.data(name='x', shape=[10], dtype='float32')
            y = fluid.layers.data(name='y', shape=[10, 20],
                             dtype='float32', lod_level=1)
Y
yangyaming 已提交
3773
            out = layers.sequence_expand(x=x, y=y, ref_level=0)
3774
    """
Y
yangyaming 已提交
3775
    helper = LayerHelper('sequence_expand', input=x, **locals())
3776
    dtype = helper.input_dtype()
X
Xin Pan 已提交
3777
    tmp = helper.create_variable_for_type_inference(dtype)
3778
    helper.append_op(
Y
yangyaming 已提交
3779 3780 3781 3782 3783
        type='sequence_expand',
        inputs={'X': x,
                'Y': y},
        outputs={'Out': tmp},
        attrs={'ref_level': ref_level})
3784
    return tmp
3785 3786


C
chengduo 已提交
3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842
def sequence_expand_as(x, y, name=None):
    """Sequence Expand As Layer. This layer will expand the input variable **x**
    according to the zeroth level lod of **y**. Current implementation requires
    the level number of Input(Y)'s lod must be 1, and the first dimension of
    Input(X) should be equal to the size of Input(Y)'s zeroth level lod, and
    lod of Input(X) is not considered.

    Following examples will explain how sequence_expand_as works:

    .. code-block:: text

        * Case 1:

            Given a 1-level LoDTensor input(X)
                X.data = [[a], [b], [c], [d]]
                X.dims = [4, 1]
            and input(Y)
                Y.lod = [[0, 3, 6, 7, 8]]
            ref_level: 0
            then we get 1-level LoDTensor
                Out.lod =  [[0,            3,              6,  7,  8]]
                Out.data = [[a], [a], [a], [b], [b], [b], [c], [d]]
                Out.dims = [8, 1]

        * Case 2:

            Given a common Tensor input(X)
                X.data = [[a, b], [c, d], [e, f]]
                X.dims = [3, 2]
            and input(Y)
                Y.lod = [[0, 2, 3, 6]]
            ref_level: 0
            then we get a common LoDTensor
                Out.lod =  [[0,             2,     3,                    6]]
                Out.data = [[a, b], [a, b] [c, d], [e, f], [e, f], [e, f]]
                Out.dims = [6, 2]

    Args:
        x (Variable): The input variable which is a Tensor or LoDTensor.
        y (Variable): The input variable which is a LoDTensor.
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.

    Returns:
        Variable: The expanded variable which is a LoDTensor.

    Examples:
        .. code-block:: python

            x = fluid.layers.data(name='x', shape=[10], dtype='float32')
            y = fluid.layers.data(name='y', shape=[10, 20],
                             dtype='float32', lod_level=1)
            out = layers.sequence_expand_as(x=x, y=y)
    """
    helper = LayerHelper('sequence_expand_as', input=x, **locals())
    dtype = helper.input_dtype()
X
Xin Pan 已提交
3843
    tmp = helper.create_variable_for_type_inference(dtype)
C
chengduo 已提交
3844 3845 3846 3847 3848 3849 3850 3851
    helper.append_op(
        type='sequence_expand_as',
        inputs={'X': x,
                'Y': y},
        outputs={'Out': tmp})
    return tmp


F
fengjiayi 已提交
3852
@templatedoc()
3853
def sequence_pad(x, pad_value, maxlen=None, name=None):
F
fengjiayi 已提交
3854 3855 3856 3857 3858
    """
    ${comment}

    Args:
        x(Variable): Input variable which should contain lod information.
M
minqiyang 已提交
3859 3860 3861
        pad_value(Variable): The Variable that holds values that will be fill
            into padded steps. It can be a scalar or a tensor whose shape
            equals to time steps in sequences. If it's a scalar, it will be
F
fengjiayi 已提交
3862
            automatically broadcasted to the shape of time step.
M
minqiyang 已提交
3863 3864 3865 3866
        maxlen(int, default None): The length of padded sequences. It can be
            None or any positive int. When it is None, all sequences will be
            padded up to the length of the longest one among them; when it a
            certain positive value, it must be greater than the length of the
3867 3868 3869
            longest original sequence.
        name(str|None): A name for this layer(optional). If set None, the layer
            will be named automatically.
M
minqiyang 已提交
3870

F
fengjiayi 已提交
3871
    Returns:
M
minqiyang 已提交
3872
        Variable: The padded sequence batch and the original lengths before
3873
                  padding. All sequences has the same length.
M
minqiyang 已提交
3874

F
fengjiayi 已提交
3875 3876 3877 3878 3879 3880 3881
    Examples:
        .. code-block:: python

            import numpy

            x = fluid.layers.data(name='y', shape=[10, 5],
                             dtype='float32', lod_level=1)
G
gmcather 已提交
3882
            pad_value = fluid.layers.assign(
D
dongzhihong 已提交
3883
                input=numpy.array([0.0], dtype=numpy.float32))
F
fengjiayi 已提交
3884 3885 3886 3887 3888
            out = fluid.layers.sequence_pad(x=x, pad_value=pad_value)
    """

    helper = LayerHelper('sequence_pad', input=x, **locals())
    dtype = helper.input_dtype()
X
Xin Pan 已提交
3889 3890
    out = helper.create_variable_for_type_inference(dtype)
    length = helper.create_variable_for_type_inference(dtype)
3891 3892 3893 3894

    pad_value.stop_gradient = True
    length.stop_gradient = True

F
fengjiayi 已提交
3895 3896 3897 3898 3899 3900
    if maxlen is None:
        maxlen = -1
    helper.append_op(
        type='sequence_pad',
        inputs={'X': x,
                'PadValue': pad_value},
3901 3902
        outputs={'Out': out,
                 'Length': length},
F
fengjiayi 已提交
3903
        attrs={'padded_length': maxlen})
3904
    return out, length
F
fengjiayi 已提交
3905 3906


3907
def sequence_unpad(x, length, name=None):
Y
Yibing Liu 已提交
3908
    """
3909
    **Sequence Unpad Layer**
Y
Yibing Liu 已提交
3910

3911 3912
    This layer removes the padding data in the input sequences and convert
    them into sequences with actual length as output, identitied by lod
Y
Yibing Liu 已提交
3913 3914 3915 3916 3917 3918 3919 3920 3921
    information.

    .. code-block:: text

	Example:

	Given input Variable **x**:
	    x.data = [[ 1.0,  2.0,  3.0,  4.0,  5.0],
		      [ 6.0,  7.0,  8.0,  9.0, 10.0],
3922 3923 3924
		      [11.0, 12.0, 13.0, 14.0, 15.0]],

	in which there are 3 sequences padded to length 5, and the acutal length
3925
	specified by input Variable **length**:
Y
Yibing Liu 已提交
3926 3927 3928 3929 3930 3931

	    length.data = [[2], [3], [4]],

	after unpadding, the output Variable will be:

	    out.data = [[1.0, 2.0, 6.0, 7.0, 8.0, 11.0, 12.0, 13.0, 14.0]]
3932
	    out.lod = [[2, 3, 4]]
Y
Yibing Liu 已提交
3933 3934 3935 3936 3937 3938

    Args:
        x(Variable): Input Variable which contains the padded sequences with
            equal length.
        length(Variable): The Variable that specifies the actual ength of
            sequences after unpadding.
3939 3940
        name(str|None): A name for this layer(optional). If set None, the layer
            will be named automatically.
Y
Yibing Liu 已提交
3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954

    Returns:
        Variable: The Variable contains the unpadded sequences.

    Examples:
        .. code-block:: python

            x = fluid.layers.data(name='x', shape=[10, 5], dtype='float32')
            len = fluid.layers.data(name='length', shape=[1], dtype='int64')
            out = fluid.layers.sequence_unpad(x=x, length=len)
    """

    helper = LayerHelper('sequence_unpad', input=x, **locals())
    dtype = helper.input_dtype()
X
Xin Pan 已提交
3955
    out = helper.create_variable_for_type_inference(dtype)
Y
Yibing Liu 已提交
3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966

    length.stop_gradient = True

    helper.append_op(
        type='sequence_unpad',
        inputs={'X': x,
                'Length': length},
        outputs={'Out': out})
    return out


3967 3968 3969 3970 3971 3972 3973
def beam_search(pre_ids,
                pre_scores,
                ids,
                scores,
                beam_size,
                end_id,
                level=0,
3974
                is_accumulated=True,
3975 3976
                name=None,
                return_parent_idx=False):
3977
    """
3978 3979
    Beam search is a classical algorithm for selecting candidate words in a
    machine translation task.
Y
Yan Chunwei 已提交
3980 3981 3982

    Refer to `Beam search <https://en.wikipedia.org/wiki/Beam_search>`_
    for more details.
M
minqiyang 已提交
3983 3984

    This layer does the search in beams for one time step. Specifically, it
3985 3986 3987
    selects the top-K candidate word ids of current step from :attr:`ids`
    according to their :attr:`scores` for all source sentences, where K is
    :attr:`beam_size` and :attr:`ids, scores` are predicted results from the
3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998
    computation cell. If :attr:`ids` is not set, it will be calculated out
    according to :attr:`scores`. Additionally, :attr:`pre_ids` and
    :attr:`pre_scores` are the output of beam_search at previous step, they
    are needed for special use to handle ended candidate translations.

    Note that if :attr:`is_accumulated` is :attr:`True`, the :attr:`scores`
    passed in should be accumulated scores. Else, the :attr:`scores` are
    considered as the straightforward scores and will be transformed to the
    log field and accumulated the :attr:`pre_scores` in this operator.
    Length penalty should be done with extra operators before calculating the
    accumulated scores if needed.
3999 4000 4001 4002

    Please see the following demo for a fully beam search usage example:

        fluid/tests/book/test_machine_translation.py
Y
Yan Chunwei 已提交
4003

4004
    Args:
4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027
        pre_ids(Variable): The LodTensor variable which is the output of
            beam_search at previous step. It should be a LodTensor with shape
            :math:`(batch_size, 1)` and lod
            :math:`[[0, 1, ... , batch_size], [0, 1, ..., batch_size]]` at the
            first step.
        pre_scores(Variable): The LodTensor variable which is the output of
            beam_search at previous step.
        ids(Variable): The LodTensor variable containing the candidates ids.
            Its shape should be :math:`(batch_size \\times beam_size, K)`,
            where :math:`K` supposed to be :attr:`beam_size`.
        scores(Variable): The LodTensor variable containing the accumulated
            scores corresponding to :attr:`ids` and its shape is the same as
            the shape of :attr:`ids`.
        beam_size(int): The beam width used in beam search.
        end_id(int): The id of end token.
        level(int, default 0): It can be ignored and mustn't change currently.
            It means the source level of lod, which is explained as following.
            The lod level of :attr:`ids` should be 2. The first level is source
            level which describes how many prefixes (branchs) for each source
            sentece (beam), and the second level is sentence level which
            describes how these candidates belong to the prefix. The paths
            linking prefixes and selected candidates are organized and reserved
            in lod.
4028 4029
        is_accumulated(bool, default True): Whether the input :attr:`score` is
             accumulated scores.
4030 4031
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.
4032 4033 4034 4035
        return_parent_idx(bool): Whether to return an extra Tensor variable 
                        preserving the selected_ids' parent indice in pre_ids
                        in output, which can be used to gather cell states at
                        the next time step.
F
fengjiayi 已提交
4036

4037
    Returns:
4038 4039 4040 4041
        Variable: The LodTensor tuple containing the selected ids and the \
            corresponding scores. If :attr:`return_parent_idx` is :attr:`True`, \
            an extra Tensor variable preserving the selected_ids' parent indice \
            is included.
Y
Yan Chunwei 已提交
4042 4043 4044 4045

    Examples:
        .. code-block:: python

4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062
            # Suppose `probs` contains predicted results from the computation
            # cell and `pre_ids` and `pre_scores` is the output of beam_search
            # at previous step.
            topk_scores, topk_indices = layers.topk(probs, k=beam_size)
            accu_scores = layers.elementwise_add(
                x=layers.log(x=topk_scores)),
                y=layers.reshape(
                    pre_scores, shape=[-1]),
                axis=0)
            selected_ids, selected_scores = layers.beam_search(
                pre_ids=pre_ids,
                pre_scores=pre_scores,
                ids=topk_indices,
                scores=accu_scores,
                beam_size=beam_size,
                end_id=end_id)
    """
Q
Qiao Longfei 已提交
4063
    helper = LayerHelper('beam_search', **locals())
4064 4065 4066 4067 4068 4069
    score_type = pre_scores.dtype
    id_type = pre_ids.dtype

    inputs = {"pre_ids": pre_ids, "pre_scores": pre_scores, "scores": scores}
    if ids is not None:
        inputs["ids"] = ids
Q
Qiao Longfei 已提交
4070

X
Xin Pan 已提交
4071 4072 4073
    selected_scores = helper.create_variable_for_type_inference(
        dtype=score_type)
    selected_ids = helper.create_variable_for_type_inference(dtype=id_type)
4074 4075 4076 4077 4078
    # parent_idx is a tensor used to gather cell states at the next time
    # step. Though lod in selected_ids can also be used to gather by
    # sequence_expand, it is not efficient.
    # gather_op's index input only supports int32 dtype currently
    parent_idx = helper.create_variable_for_type_inference(dtype="int32")
Q
Qiao Longfei 已提交
4079 4080 4081

    helper.append_op(
        type='beam_search',
4082
        inputs=inputs,
Q
Qiao Longfei 已提交
4083 4084 4085
        outputs={
            'selected_ids': selected_ids,
            'selected_scores': selected_scores,
4086
            'parent_idx': parent_idx
Q
Qiao Longfei 已提交
4087 4088 4089 4090 4091 4092
        },
        attrs={
            # TODO(ChunweiYan) to assure other value support
            'level': level,
            'beam_size': beam_size,
            'end_id': end_id,
4093
            'is_accumulated': is_accumulated,
Q
Qiao Longfei 已提交
4094
        })
4095 4096 4097 4098
    if return_parent_idx:
        return selected_ids, selected_scores, parent_idx
    else:
        return selected_ids, selected_scores
Q
Qiao Longfei 已提交
4099 4100


4101 4102 4103 4104 4105 4106 4107
def beam_search_decode(ids, scores, beam_size, end_id, name=None):
    """
    Beam Search Decode Layer. This layer constructs the full hypotheses for
    each source sentence by walking back along the LoDTensorArray :attr:`ids`
    whose lods can be used to restore the path in the beam search tree.
    Please see the following demo for a fully beam search usage example:
        fluid/tests/book/test_machine_translation.py
G
guosheng 已提交
4108

4109 4110 4111 4112 4113 4114 4115 4116 4117
    Args:
        ids(Variable): The LodTensorArray variable containing the selected ids
            of all steps.
        scores(Variable): The LodTensorArray variable containing the selected
            scores of all steps.
        beam_size(int): The beam width used in beam search.
        end_id(int): The id of end token.
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.
G
guosheng 已提交
4118

4119 4120 4121 4122 4123 4124
    Returns:
        Variable: The LodTensor pair containing the generated id sequences \
            and the corresponding scores. The shapes and lods of the two \
            LodTensor are same. The lod level is 2 and the two levels \
            separately indicate how many hypotheses each source sentence has \
            and how many ids each hypothesis has.
G
guosheng 已提交
4125

4126 4127
    Examples:
        .. code-block:: python
T
Tink_Y 已提交
4128

4129 4130 4131 4132 4133 4134
            # Suppose `ids` and `scores` are LodTensorArray variables reserving
            # the selected ids and scores of all steps
            finished_ids, finished_scores = layers.beam_search_decode(
                ids, scores, beam_size=5, end_id=0)
    """
    helper = LayerHelper('beam_search_decode', **locals())
X
Xin Pan 已提交
4135 4136
    sentence_ids = helper.create_variable_for_type_inference(dtype=ids.dtype)
    sentence_scores = helper.create_variable_for_type_inference(dtype=ids.dtype)
4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151

    helper.append_op(
        type="beam_search_decode",
        inputs={"Ids": ids,
                "Scores": scores},
        outputs={
            "SentenceIds": sentence_ids,
            "SentenceScores": sentence_scores
        },
        attrs={"beam_size": beam_size,
               "end_id": end_id})

    return sentence_ids, sentence_scores


Y
yangyaming 已提交
4152 4153 4154 4155
def lstm_unit(x_t,
              hidden_t_prev,
              cell_t_prev,
              forget_bias=0.0,
Y
yangyaming 已提交
4156
              param_attr=None,
C
caoying03 已提交
4157 4158
              bias_attr=None,
              name=None):
Y
yangyaming 已提交
4159 4160 4161 4162
    """Lstm unit layer. The equation of a lstm step is:

        .. math::

4163
            i_t & = \sigma(W_{x_i}x_{t} + W_{h_i}h_{t-1} + b_i)
Y
yangyaming 已提交
4164

4165
            f_t & = \sigma(W_{x_f}x_{t} + W_{h_f}h_{t-1} + b_f)
Y
yangyaming 已提交
4166

4167
            c_t & = f_tc_{t-1} + i_t tanh (W_{x_c}x_t + W_{h_c}h_{t-1} + b_c)
Y
yangyaming 已提交
4168

4169
            o_t & = \sigma(W_{x_o}x_{t} + W_{h_o}h_{t-1} + b_o)
Y
yangyaming 已提交
4170 4171 4172

            h_t & = o_t tanh(c_t)

4173 4174 4175 4176 4177 4178
    The inputs of lstm unit include :math:`x_t`, :math:`h_{t-1}` and
    :math:`c_{t-1}`. The 2nd dimensions of :math:`h_{t-1}` and :math:`c_{t-1}`
    should be same. The implementation separates the linear transformation and
    non-linear transformation apart. Here, we take :math:`i_t` as an example.
    The linear transformation is applied by calling a `fc` layer and the
    equation is:
Y
yangyaming 已提交
4179 4180 4181

        .. math::

4182
            L_{i_t} = W_{x_i}x_{t} + W_{h_i}h_{t-1} + b_i
Y
yangyaming 已提交
4183 4184 4185 4186 4187 4188 4189 4190

    The non-linear transformation is applied by calling `lstm_unit_op` and the
    equation is:

        .. math::

            i_t = \sigma(L_{i_t})

Y
yangyaming 已提交
4191
    This layer has two outputs including :math:`h_t` and :math:`o_t`.
Y
yangyaming 已提交
4192 4193

    Args:
Y
yangyaming 已提交
4194 4195 4196 4197 4198 4199
        x_t (Variable): The input value of current step, a 2-D tensor with shape
            M x N, M for batch size and N for input size.
        hidden_t_prev (Variable): The hidden value of lstm unit, a 2-D tensor
            with shape M x S, M for batch size and S for size of lstm unit.
        cell_t_prev (Variable): The cell value of lstm unit, a 2-D tensor with
            shape M x S, M for batch size and S for size of lstm unit.
Y
yangyaming 已提交
4200
        forget_bias (float): The forget bias of lstm unit.
C
chengduo 已提交
4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212
        param_attr(ParamAttr|None): The parameter attribute for the learnable
                               hidden-hidden weights.
                               If it is set to None or one attribute of ParamAttr,
                               lstm_unit will create ParamAttr as param_attr.
                               If the Initializer of the param_attr is not set, the
                               parameter is initialized with Xavier. Default: None.
        bias_attr (ParamAttr|None): The bias attribute for the learnable bias
                              weights. If it is set to False, no bias will be added
                              to the output units. If it is set to None or one attribute of ParamAttr,
                              lstm_unit will create ParamAttr as bias_attr.
                              If the Initializer of the bias_attr is not set,
                              the bias is initialized zero. Default: None.
C
caoying03 已提交
4213 4214
        name(str|None): A name for this layer(optional). If set None, the layer
                       will be named automatically.
Y
yangyaming 已提交
4215 4216

    Returns:
Y
yangyaming 已提交
4217
        tuple: The hidden value and cell value of lstm unit.
Y
yangyaming 已提交
4218 4219

    Raises:
4220 4221 4222 4223
        ValueError: The ranks of **x_t**, **hidden_t_prev** and **cell_t_prev**
                    not be 2 or the 1st dimensions of **x_t**, **hidden_t_prev**
                    and **cell_t_prev** not be the same or the 2nd dimensions of
                    **hidden_t_prev** and **cell_t_prev** not be the same.
Y
yangyaming 已提交
4224 4225 4226 4227 4228 4229

    Examples:

        .. code-block:: python

             x_t = fluid.layers.fc(input=x_t_data, size=10)
4230
             prev_hidden = fluid.layers.fc(input=prev_hidden_data, size=30)
Y
yangyaming 已提交
4231
             prev_cell = fluid.layers.fc(input=prev_cell_data, size=30)
Y
yangyaming 已提交
4232
             hidden_value, cell_value = fluid.layers.lstm_unit(x_t=x_t,
Y
yangyaming 已提交
4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248
                                                    hidden_t_prev=prev_hidden,
                                                    cell_t_prev=prev_cell)
    """
    helper = LayerHelper('lstm_unit', **locals())

    if len(x_t.shape) != 2:
        raise ValueError("Rank of x_t must be 2.")

    if len(hidden_t_prev.shape) != 2:
        raise ValueError("Rank of hidden_t_prev must be 2.")

    if len(cell_t_prev.shape) != 2:
        raise ValueError("Rank of cell_t_prev must be 2.")

    if x_t.shape[0] != hidden_t_prev.shape[0] or x_t.shape[
            0] != cell_t_prev.shape[0]:
Y
yangyaming 已提交
4249
        raise ValueError("The 1st dimensions of x_t, hidden_t_prev and "
4250 4251 4252 4253
                         "cell_t_prev must be the same.")

    if hidden_t_prev.shape[1] != cell_t_prev.shape[1]:
        raise ValueError("The 2nd dimensions of hidden_t_prev and "
Y
yangyaming 已提交
4254 4255
                         "cell_t_prev must be the same.")

Y
yangyaming 已提交
4256 4257 4258
    if bias_attr is None:
        bias_attr = ParamAttr()

Y
yangyaming 已提交
4259
    size = cell_t_prev.shape[1]
4260
    concat_out = concat(input=[x_t, hidden_t_prev], axis=1)
Y
yangyaming 已提交
4261 4262
    fc_out = fc(input=concat_out,
                size=4 * size,
Y
yangyaming 已提交
4263
                param_attr=param_attr,
4264
                bias_attr=bias_attr)
Y
yangyaming 已提交
4265
    dtype = x_t.dtype
X
Xin Pan 已提交
4266 4267
    c = helper.create_variable_for_type_inference(dtype)
    h = helper.create_variable_for_type_inference(dtype)
Y
yangyaming 已提交
4268 4269 4270 4271 4272 4273 4274 4275 4276

    helper.append_op(
        type='lstm_unit',
        inputs={"X": fc_out,
                "C_prev": cell_t_prev},
        outputs={"C": c,
                 "H": h},
        attrs={"forget_bias": forget_bias})

Y
yangyaming 已提交
4277
    return h, c
G
guosheng 已提交
4278 4279


C
caoying03 已提交
4280
def reduce_sum(input, dim=None, keep_dim=False, name=None):
G
guosheng 已提交
4281
    """
Y
yangyaming 已提交
4282
    Computes the sum of tensor elements over the given dimension.
G
guosheng 已提交
4283 4284 4285

    Args:
        input (Variable): The input variable which is a Tensor or LoDTensor.
W
whs 已提交
4286
        dim (list|int|None): The dimensions along which the sum is performed. If
Y
yangyaming 已提交
4287 4288
            :attr:`None`, sum all elements of :attr:`input` and return a
            Tensor variable with a single element, otherwise must be in the
W
whs 已提交
4289 4290
            range :math:`[-rank(input), rank(input))`. If :math:`dim[i] < 0`,
            the dimension to reduce is :math:`rank + dim[i]`.
4291
        keep_dim (bool|False): Whether to reserve the reduced dimension in the
Y
yangyaming 已提交
4292
            output Tensor. The result tensor will have one fewer dimension
G
guosheng 已提交
4293
            than the :attr:`input` unless :attr:`keep_dim` is true.
C
caoying03 已提交
4294 4295
        name(str|None): A name for this layer(optional). If set None, the layer
                       will be named automatically.
G
guosheng 已提交
4296 4297 4298

    Returns:
        Variable: The reduced Tensor variable.
F
fengjiayi 已提交
4299

G
guosheng 已提交
4300 4301 4302 4303 4304 4305
    Examples:
        .. code-block:: python

            # x is a Tensor variable with following elements:
            #    [[0.2, 0.3, 0.5, 0.9]
            #     [0.1, 0.2, 0.6, 0.7]]
Q
qiaolongfei 已提交
4306
            # Each example is followed by the corresponding output tensor.
G
guosheng 已提交
4307 4308 4309 4310
            fluid.layers.reduce_sum(x)  # [3.5]
            fluid.layers.reduce_sum(x, dim=0)  # [0.3, 0.5, 1.1, 1.6]
            fluid.layers.reduce_sum(x, dim=-1)  # [1.9, 1.6]
            fluid.layers.reduce_sum(x, dim=1, keep_dim=True)  # [[1.9], [1.6]]
W
whs 已提交
4311 4312 4313 4314

            # x is a Tensor variable with shape [2, 2, 2] and elements as below:
            #      [[[1, 2], [3, 4]],
            #      [[5, 6], [7, 8]]]
Q
qiaolongfei 已提交
4315
            # Each example is followed by the corresponding output tensor.
W
whs 已提交
4316 4317 4318
            fluid.layers.reduce_sum(x, dim=[1, 2]) # [10, 26]
            fluid.layers.reduce_sum(x, dim=[0, 1]) # [16, 20]

G
guosheng 已提交
4319 4320
    """
    helper = LayerHelper('reduce_sum', **locals())
X
Xin Pan 已提交
4321
    out = helper.create_variable_for_type_inference(dtype=helper.input_dtype())
W
whs 已提交
4322 4323
    if dim is not None and not isinstance(dim, list):
        dim = [dim]
G
guosheng 已提交
4324 4325 4326 4327 4328
    helper.append_op(
        type='reduce_sum',
        inputs={'X': input},
        outputs={'Out': out},
        attrs={
W
whs 已提交
4329
            'dim': dim if dim != None else [0],
G
guosheng 已提交
4330 4331 4332 4333
            'keep_dim': keep_dim,
            'reduce_all': True if dim == None else False
        })
    return out
G
guosheng 已提交
4334 4335


C
caoying03 已提交
4336
def reduce_mean(input, dim=None, keep_dim=False, name=None):
G
guosheng 已提交
4337
    """
Y
Yibing Liu 已提交
4338
    Computes the mean of the input tensor's elements along the given dimension.
G
guosheng 已提交
4339 4340 4341

    Args:
        input (Variable): The input variable which is a Tensor or LoDTensor.
Y
Yibing Liu 已提交
4342 4343 4344
        dim (list|int|None): The dimension along which the mean is computed. If
            `None`, compute the mean over all elements of :attr:`input`
            and return a variable with a single element, otherwise it
Y
yangyaming 已提交
4345
            must be in the range :math:`[-rank(input), rank(input))`. If
4346
            :math:`dim[i] < 0`, the dimension to reduce is
Y
Yibing Liu 已提交
4347
            :math:`rank(input) + dim[i]`.
Y
yangyaming 已提交
4348 4349
        keep_dim (bool): Whether to reserve the reduced dimension in the
            output Tensor. The result tensor will have one fewer dimension
G
guosheng 已提交
4350
            than the :attr:`input` unless :attr:`keep_dim` is true.
Y
Yibing Liu 已提交
4351
        name(str|None): A name for this layer(optional). If set `None`, the layer
C
caoying03 已提交
4352
                       will be named automatically.
G
guosheng 已提交
4353 4354

    Returns:
Y
Yibing Liu 已提交
4355
        Variable: The reduced mean Variable.
F
fengjiayi 已提交
4356

G
guosheng 已提交
4357 4358 4359 4360 4361 4362 4363 4364 4365 4366
    Examples:
        .. code-block:: python

            # x is a Tensor variable with following elements:
            #    [[0.2, 0.3, 0.5, 0.9]
            #     [0.1, 0.2, 0.6, 0.7]]
            # Each example is followed by the correspending output tensor.
            fluid.layers.reduce_mean(x)  # [0.4375]
            fluid.layers.reduce_mean(x, dim=0)  # [0.15, 0.25, 0.55, 0.8]
            fluid.layers.reduce_mean(x, dim=-1)  # [0.475, 0.4]
F
stash  
fengjiayi 已提交
4367 4368
            fluid.layers.reduce_mean(
                x, dim=1, keep_dim=True)  # [[0.475], [0.4]]
W
whs 已提交
4369 4370 4371 4372 4373 4374 4375

            # x is a Tensor variable with shape [2, 2, 2] and elements as below:
            #      [[[1.0, 2.0], [3.0, 4.0]],
            #      [[5.0, 6.0], [7.0, 8.0]]]
            # Each example is followed by the correspending output tensor.
            fluid.layers.reduce_mean(x, dim=[1, 2]) # [2.5, 6.5]
            fluid.layers.reduce_mean(x, dim=[0, 1]) # [4.0, 5.0]
G
guosheng 已提交
4376 4377
    """
    helper = LayerHelper('reduce_mean', **locals())
X
Xin Pan 已提交
4378
    out = helper.create_variable_for_type_inference(dtype=helper.input_dtype())
W
whs 已提交
4379 4380
    if dim is not None and not isinstance(dim, list):
        dim = [dim]
G
guosheng 已提交
4381 4382 4383 4384 4385
    helper.append_op(
        type='reduce_mean',
        inputs={'X': input},
        outputs={'Out': out},
        attrs={
W
whs 已提交
4386
            'dim': dim if dim != None else [0],
G
guosheng 已提交
4387 4388 4389 4390
            'keep_dim': keep_dim,
            'reduce_all': True if dim == None else False
        })
    return out
4391 4392


C
caoying03 已提交
4393
def reduce_max(input, dim=None, keep_dim=False, name=None):
4394
    """
Y
yangyaming 已提交
4395
    Computes the maximum of tensor elements over the given dimension.
4396 4397 4398

    Args:
        input (Variable): The input variable which is a Tensor or LoDTensor.
W
whs 已提交
4399
        dim (list|int|None): The dimension along which the maximum is computed.
Y
yangyaming 已提交
4400 4401 4402
            If :attr:`None`, compute the maximum over all elements of
            :attr:`input` and return a Tensor variable with a single element,
            otherwise must be in the range :math:`[-rank(input), rank(input))`.
W
whs 已提交
4403
            If :math:`dim[i] < 0`, the dimension to reduce is :math:`rank + dim[i]`.
Y
yangyaming 已提交
4404 4405
        keep_dim (bool): Whether to reserve the reduced dimension in the
            output Tensor. The result tensor will have one fewer dimension
4406
            than the :attr:`input` unless :attr:`keep_dim` is true.
C
caoying03 已提交
4407 4408
        name(str|None): A name for this layer(optional). If set None, the layer
                       will be named automatically.
4409 4410 4411

    Returns:
        Variable: The reduced Tensor variable.
Y
yangyaming 已提交
4412

4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423
    Examples:
        .. code-block:: python

            # x is a Tensor variable with following elements:
            #    [[0.2, 0.3, 0.5, 0.9]
            #     [0.1, 0.2, 0.6, 0.7]]
            # Each example is followed by the correspending output tensor.
            fluid.layers.reduce_max(x)  # [0.9]
            fluid.layers.reduce_max(x, dim=0)  # [0.2, 0.3, 0.6, 0.9]
            fluid.layers.reduce_max(x, dim=-1)  # [0.9, 0.7]
            fluid.layers.reduce_max(x, dim=1, keep_dim=True)  # [[0.9], [0.7]]
W
whs 已提交
4424 4425 4426 4427 4428 4429 4430

            # x is a Tensor variable with shape [2, 2, 2] and elements as below:
            #      [[[1.0, 2.0], [3.0, 4.0]],
            #      [[5.0, 6.0], [7.0, 8.0]]]
            # Each example is followed by the correspending output tensor.
            fluid.layers.reduce_max(x, dim=[1, 2]) # [4.0, 8.0]
            fluid.layers.reduce_max(x, dim=[0, 1]) # [7.0, 8.0]
4431 4432
    """
    helper = LayerHelper('reduce_max', **locals())
X
Xin Pan 已提交
4433
    out = helper.create_variable_for_type_inference(dtype=helper.input_dtype())
W
whs 已提交
4434 4435
    if dim is not None and not isinstance(dim, list):
        dim = [dim]
4436 4437 4438 4439 4440
    helper.append_op(
        type='reduce_max',
        inputs={'X': input},
        outputs={'Out': out},
        attrs={
W
whs 已提交
4441
            'dim': dim if dim != None else [0],
4442 4443 4444 4445 4446 4447
            'keep_dim': keep_dim,
            'reduce_all': True if dim == None else False
        })
    return out


C
caoying03 已提交
4448
def reduce_min(input, dim=None, keep_dim=False, name=None):
4449
    """
Y
yangyaming 已提交
4450
    Computes the minimum of tensor elements over the given dimension.
4451 4452 4453

    Args:
        input (Variable): The input variable which is a Tensor or LoDTensor.
W
whs 已提交
4454
        dim (list|int|None): The dimensions along which the minimum is computed.
Y
yangyaming 已提交
4455 4456 4457
            If :attr:`None`, compute the minimum over all elements of
            :attr:`input` and return a Tensor variable with a single element,
            otherwise must be in the range :math:`[-rank(input), rank(input))`.
W
whs 已提交
4458
            If :math:`dim[i] < 0`, the dimension to reduce is :math:`rank + dim[i]`.
Y
yangyaming 已提交
4459 4460
        keep_dim (bool): Whether to reserve the reduced dimension in the
            output Tensor. The result tensor will have one fewer dimension
4461
            than the :attr:`input` unless :attr:`keep_dim` is true.
C
caoying03 已提交
4462 4463
        name(str|None): A name for this layer(optional). If set None, the layer
                       will be named automatically.
4464 4465 4466

    Returns:
        Variable: The reduced Tensor variable.
Y
yangyaming 已提交
4467

4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478
    Examples:
        .. code-block:: python

            # x is a Tensor variable with following elements:
            #    [[0.2, 0.3, 0.5, 0.9]
            #     [0.1, 0.2, 0.6, 0.7]]
            # Each example is followed by the correspending output tensor.
            fluid.layers.reduce_min(x)  # [0.1]
            fluid.layers.reduce_min(x, dim=0)  # [0.1, 0.2, 0.5, 0.7]
            fluid.layers.reduce_min(x, dim=-1)  # [0.2, 0.1]
            fluid.layers.reduce_min(x, dim=1, keep_dim=True)  # [[0.2], [0.1]]
W
whs 已提交
4479 4480 4481 4482 4483 4484 4485

            # x is a Tensor variable with shape [2, 2, 2] and elements as below:
            #      [[[1.0, 2.0], [3.0, 4.0]],
            #      [[5.0, 6.0], [7.0, 8.0]]]
            # Each example is followed by the correspending output tensor.
            fluid.layers.reduce_min(x, dim=[1, 2]) # [1.0, 5.0]
            fluid.layers.reduce_min(x, dim=[0, 1]) # [1.0, 2.0]
4486 4487
    """
    helper = LayerHelper('reduce_min', **locals())
X
Xin Pan 已提交
4488
    out = helper.create_variable_for_type_inference(dtype=helper.input_dtype())
W
whs 已提交
4489 4490
    if dim is not None and not isinstance(dim, list):
        dim = [dim]
4491 4492 4493 4494 4495
    helper.append_op(
        type='reduce_min',
        inputs={'X': input},
        outputs={'Out': out},
        attrs={
W
whs 已提交
4496
            'dim': dim if dim != None else [0],
4497 4498 4499 4500
            'keep_dim': keep_dim,
            'reduce_all': True if dim == None else False
        })
    return out
G
guosheng 已提交
4501 4502


4503 4504 4505 4506 4507 4508
def reduce_prod(input, dim=None, keep_dim=False, name=None):
    """
    Computes the product of tensor elements over the given dimension.

    Args:
        input (Variable): The input variable which is a Tensor or LoDTensor.
W
whs 已提交
4509
        dim (list|int|None): The dimensions along which the product is performed. If
4510 4511
            :attr:`None`, multipy all elements of :attr:`input` and return a
            Tensor variable with a single element, otherwise must be in the
W
whs 已提交
4512 4513
            range :math:`[-rank(input), rank(input))`. If :math:`dim[i] < 0`,
            the dimension to reduce is :math:`rank + dim[i]`.
4514 4515 4516
        keep_dim (bool|False): Whether to reserve the reduced dimension in the
            output Tensor. The result tensor will have one fewer dimension
            than the :attr:`input` unless :attr:`keep_dim` is true.
Y
yangyaming 已提交
4517
        name(str|None): A name for this layer(optional). If set None, the
Z
zhouhanqing 已提交
4518
            layer will be named automatically.
4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532

    Returns:
        Variable: The reduced Tensor variable.

    Examples:
        .. code-block:: python

            # x is a Tensor variable with following elements:
            #    [[0.2, 0.3, 0.5, 0.9]
            #     [0.1, 0.2, 0.6, 0.7]]
            # Each example is followed by the correspending output tensor.
            fluid.layers.reduce_prod(x)  # [0.0002268]
            fluid.layers.reduce_prod(x, dim=0)  # [0.02, 0.06, 0.3, 0.63]
            fluid.layers.reduce_prod(x, dim=-1)  # [0.027, 0.0084]
Y
yangyaming 已提交
4533
            fluid.layers.reduce_prod(x, dim=1,
Z
zhouhanqing 已提交
4534
                                     keep_dim=True)  # [[0.027], [0.0084]]
W
whs 已提交
4535 4536 4537 4538 4539 4540 4541

            # x is a Tensor variable with shape [2, 2, 2] and elements as below:
            #      [[[1.0, 2.0], [3.0, 4.0]],
            #      [[5.0, 6.0], [7.0, 8.0]]]
            # Each example is followed by the correspending output tensor.
            fluid.layers.reduce_prod(x, dim=[1, 2]) # [24.0, 1680.0]
            fluid.layers.reduce_prod(x, dim=[0, 1]) # [105.0, 384.0]
4542 4543
    """
    helper = LayerHelper('reduce_prod', **locals())
X
Xin Pan 已提交
4544
    out = helper.create_variable_for_type_inference(dtype=helper.input_dtype())
W
whs 已提交
4545 4546
    if dim is not None and not isinstance(dim, list):
        dim = [dim]
4547 4548 4549 4550 4551
    helper.append_op(
        type='reduce_prod',
        inputs={'X': input},
        outputs={'Out': out},
        attrs={
W
whs 已提交
4552
            'dim': dim if dim != None else [0],
4553 4554 4555 4556 4557 4558
            'keep_dim': keep_dim,
            'reduce_all': True if dim == None else False
        })
    return out


C
caoying03 已提交
4559
def split(input, num_or_sections, dim=-1, name=None):
G
guosheng 已提交
4560
    """
C
caoying03 已提交
4561
    Split the input tensor into multiple sub-tensors.
G
guosheng 已提交
4562 4563 4564

    Args:
        input (Variable): The input variable which is a Tensor or LoDTensor.
C
caoying03 已提交
4565 4566 4567 4568 4569
        num_or_sections (int|list): If :attr:`num_or_sections` is an integer,
            then the integer indicates the number of equal sized sub-tensors
            that the tensor will be divided into. If :attr:`num_or_sections`
            is a list of integers, the length of list indicates the number of
            sub-tensors and the integers indicate the sizes of sub-tensors'
G
guosheng 已提交
4570
            :attr:`dim` dimension orderly.
C
caoying03 已提交
4571
        dim (int): The dimension along which to split. If :math:`dim < 0`, the
G
guosheng 已提交
4572
            dimension to split along is :math:`rank(input) + dim`.
C
caoying03 已提交
4573 4574
        name(str|None): A name for this layer(optional). If set None, the layer
                       will be named automatically.
G
guosheng 已提交
4575 4576

    Returns:
D
dzhwinter 已提交
4577
        list(Variable): The list of segmented tensor variables.
G
guosheng 已提交
4578 4579 4580 4581 4582 4583 4584 4585 4586

    Examples:
        .. code-block:: python

            # x is a Tensor variable with shape [3, 9, 5]:
            x0, x1, x2 = fluid.layers.split(x, num_or_sections=3, dim=1)
            x0.shape  # [3, 3, 5]
            x1.shape  # [3, 3, 5]
            x2.shape  # [3, 3, 5]
F
stash  
fengjiayi 已提交
4587 4588
            x0, x1, x2 = fluid.layers.split(
                x, num_or_sections=[2, 3, 4], dim=1)
G
guosheng 已提交
4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603
            x0.shape  # [3, 2, 5]
            x1.shape  # [3, 3, 5]
            x2.shape  # [3, 4, 5]
    """
    helper = LayerHelper('split', **locals())
    input_shape = input.shape
    dim = (len(input_shape) + dim) if dim < 0 else dim
    if isinstance(num_or_sections, int):
        assert num_or_sections > 1, 'num_or_sections must be more than 1.'
        num = num_or_sections
    else:
        assert len(num_or_sections) < input_shape[
            dim], 'len(num_or_sections) must not be more than input.shape[dim].'
        num = len(num_or_sections)
    outs = [
X
Xin Pan 已提交
4604
        helper.create_variable_for_type_inference(dtype=helper.input_dtype())
G
guosheng 已提交
4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617
        for i in range(num)
    ]
    helper.append_op(
        type='split',
        inputs={'X': input},
        outputs={'Out': outs},
        attrs={
            'num': num_or_sections if isinstance(num_or_sections, int) else 0,
            'sections': num_or_sections
            if isinstance(num_or_sections, list) else [],
            'axis': dim
        })
    return outs
C
caoying03 已提交
4618 4619 4620 4621 4622 4623 4624 4625 4626


def l2_normalize(x, axis, epsilon=1e-12, name=None):
    """
    **L2 normalize Layer**

    The l2 normalize layer normalizes `x` along dimension `axis` using an L2
    norm. For a 1-D tensor (`dim` is fixed to 0), this layer computes

4627
    .. math::
4628 4629

        y = \\frac{x}{ \sqrt{\sum {x^2} + epsion }}
C
caoying03 已提交
4630 4631 4632 4633 4634

    For `x` with more dimensions, this layer independently normalizes each 1-D
    slice along dimension `axis`.

    Args:
4635
        x(Variable|list): The input tensor to l2_normalize layer.
4636
        axis(int): The axis on which to apply normalization. If `axis < 0`, \
4637 4638
            the dimension to normalization is rank(X) + axis. -1 is the
            last dimension.
4639
        epsilon(float): The epsilon value is used to avoid division by zero, \
4640
            the defalut value is 1e-10.
4641
        name(str|None): A name for this layer(optional). If set None, the layer \
4642
            will be named automatically.
C
caoying03 已提交
4643 4644

    Returns:
4645
        Variable: The output tensor variable is the same shape with `x`.
C
caoying03 已提交
4646 4647

    Examples:
4648

C
caoying03 已提交
4649 4650
        .. code-block:: python

4651 4652 4653 4654
            data = fluid.layers.data(name="data",
                                     shape=(3, 17, 13),
                                     dtype="float32")
            normed = fluid.layers.l2_normalize(x=data, axis=1)
C
caoying03 已提交
4655 4656
    """

F
fengjiayi 已提交
4657 4658
    if len(x.shape) == 1:
        axis = 0
C
caoying03 已提交
4659 4660
    helper = LayerHelper("l2_normalize", **locals())

X
Xin Pan 已提交
4661 4662
    out = helper.create_variable_for_type_inference(dtype=x.dtype)
    norm = helper.create_variable_for_type_inference(dtype=x.dtype)
C
caoying03 已提交
4663
    helper.append_op(
4664 4665 4666 4667
        type="norm",
        inputs={"X": x},
        outputs={"Out": out,
                 "Norm": norm},
C
caoying03 已提交
4668
        attrs={
4669 4670
            "axis": 1 if axis is None else axis,
            "epsilon": epsilon,
C
caoying03 已提交
4671 4672
        })
    return out
4673 4674


S
sneaxiy 已提交
4675
def matmul(x, y, transpose_x=False, transpose_y=False, alpha=1.0, name=None):
G
guosheng 已提交
4676
    """
Y
ying 已提交
4677 4678 4679 4680
    Applies matrix multiplication to two tensors.

    Currently, the input tensors' rank can be any, but when the rank of any
    inputs is bigger than 3, this two inputs' rank should be equal.
G
guosheng 已提交
4681

C
chengduoZH 已提交
4682
    The actual behavior depends on the shapes of :math:`x`, :math:`y` and the
4683
    flag values of :attr:`transpose_x`, :attr:`transpose_y`. Specifically:
G
guosheng 已提交
4684

4685 4686 4687 4688 4689
    - If a transpose flag is specified, the last two dimensions of the tensor
      are transposed. If the tensor is rank-1 of shape :math:`[D]`, then for
      :math:`x` it is treated as :math:`[1, D]` in nontransposed form and as
      :math:`[D, 1]` in transposed form, whereas for :math:`y` it is the
      opposite: It is treated as :math:`[D, 1]` in nontransposed form and as
4690
      :math:`[1, D]` in transposed form.
G
guosheng 已提交
4691

C
chengduoZH 已提交
4692
    - After transpose, the two tensors are 2-D or n-D and matrix multiplication
4693
      performs in the following way.
G
guosheng 已提交
4694

4695
      - If both are 2-D, they are multiplied like conventional matrices.
C
chengduoZH 已提交
4696
      - If either is n-D, it is treated as a stack of matrices residing in the
Y
ying 已提交
4697
        last two dimensions and a batched matrix multiply supporting broadcast
4698
        applies on the two tensors.
G
guosheng 已提交
4699

Y
ying 已提交
4700 4701
    Also note that if the raw tensor :math:`x` or :math:`y` is rank-1 and
    nontransposed, the prepended or appended dimension :math:`1` will be
C
chengduoZH 已提交
4702
    removed after matrix multiplication.
G
guosheng 已提交
4703 4704 4705

    Args:
        x (Variable): The input variable which is a Tensor or LoDTensor.
4706 4707 4708
        y (Variable): The input variable which is a Tensor or LoDTensor.
        transpose_x (bool): Whether to transpose :math:`x` before multiplication.
        transpose_y (bool): Whether to transpose :math:`y` before multiplication.
S
sneaxiy 已提交
4709
        alpha (float): The scale of output. Default 1.0.
4710
        name(str|None): A name for this layer(optional). If set None, the layer
4711
            will be named automatically.
G
guosheng 已提交
4712 4713

    Returns:
4714
        Variable: The product Tensor variable.
G
guosheng 已提交
4715

G
guosheng 已提交
4716 4717 4718
    Examples:
        .. code-block:: python

4719
            # Examples to clarify shapes of the inputs and output
C
chengduoZH 已提交
4720 4721
            # x: [B, ..., M, K], y: [B, ..., K, N]
            fluid.layers.matmul(x, y)  # out: [B, ..., M, N]
Y
ying 已提交
4722

4723 4724
            # x: [B, M, K], y: [B, K, N]
            fluid.layers.matmul(x, y)  # out: [B, M, N]
Y
ying 已提交
4725

4726 4727
            # x: [B, M, K], y: [K, N]
            fluid.layers.matmul(x, y)  # out: [B, M, N]
Y
ying 已提交
4728

4729 4730
            # x: [M, K], y: [K, N]
            fluid.layers.matmul(x, y)  # out: [M, N]
Y
ying 已提交
4731 4732 4733 4734

            # x: [B, M, K], y: [K]
            fluid.layers.matmul(x, y)  # out: [B, M]

4735 4736
            # x: [K], y: [K]
            fluid.layers.matmul(x, y)  # out: [1]
4737

Y
ying 已提交
4738
            # x: [M], y: [N]
4739
            fluid.layers.matmul(x, y, True, True)  # out: [M, N]
G
guosheng 已提交
4740
    """
Y
ying 已提交
4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752

    def __check_input(x, y):
        if len(y.shape) > len(x.shape):
            raise ValueError(
                "Invalid inputs for matmul. "
                "x's rank should be always greater than or equal to y'rank.")

        x_shape = list(x.shape)
        y_shape = list(y.shape)
        if len(x_shape) == 1:
            x_shape = [1] + x_shape
        if len(y_shape) == 1:
Y
ying 已提交
4753
            y_shape = y_shape + [1]
Y
ying 已提交
4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769

        # check the inner 2 dimensions
        if transpose_x:
            x_shape[-2], x_shape[-1] = x_shape[-1], x_shape[-2]
        if transpose_y:
            y_shape[-2], y_shape[-1] = y_shape[-1], y_shape[-2]
        if x_shape[-1] != y_shape[-2]:
            raise ValueError("Invalid inputs for matmul.")

        if len(y_shape) > 2:
            for i, dim_x in enumerate(x_shape[:-2]):
                if dim_x != y_shape[i]:
                    raise ValueError("Invalid inputs for matmul.")

    __check_input(x, y)

4770
    helper = LayerHelper('matmul', **locals())
X
Xin Pan 已提交
4771
    out = helper.create_variable_for_type_inference(dtype=x.dtype)
G
guosheng 已提交
4772
    helper.append_op(
4773 4774 4775 4776
        type='matmul',
        inputs={'X': x,
                'Y': y},
        outputs={'Out': out},
S
sneaxiy 已提交
4777 4778 4779
        attrs={
            'transpose_X': transpose_x,
            'transpose_Y': transpose_y,
S
sneaxiy 已提交
4780
            'alpha': float(alpha),
S
sneaxiy 已提交
4781
        })
4782
    return out
4783 4784


4785
def topk(input, k, name=None):
Q
qingqing01 已提交
4786 4787 4788 4789
    """
    This operator is used to find values and indices of the k largest entries
    for the last dimension.

F
fengjiayi 已提交
4790
    If the input is a vector (1-D Tensor), finds the k largest entries in the vector
Q
qingqing01 已提交
4791 4792 4793 4794 4795 4796
    and outputs their values and indices as vectors. Thus values[j] is the j-th
    largest entry in input, and its index is indices[j].

    If the input is a Tensor with higher rank, this operator computes the top k
    entries along the last dimension.

F
fengjiayi 已提交
4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817
    For example:

    .. code-block:: text

        If:
            input = [[5, 4, 2, 3],
                     [9, 7, 10, 25],
                     [6, 2, 10, 1]]
            k = 2

        Then:
            The first output:
            values = [[5, 4],
                      [10, 25],
                      [6, 10]]

            The second output:
            indices = [[0, 1],
                       [2, 3],
                       [0, 2]]

Q
qingqing01 已提交
4818 4819 4820
    Args:
        input(Variable): The input variable which can be a vector or Tensor with
            higher rank.
W
whs 已提交
4821
        k(int | Variable):  The number of top elements to look for along the last dimension
F
fengjiayi 已提交
4822
                 of input.
4823
        name(str|None): A name for this layer(optional). If set None, the layer
4824
                       will be named automatically.
F
fengjiayi 已提交
4825
                       Default: None
Q
qingqing01 已提交
4826 4827

    Returns:
4828 4829 4830
        Tuple[Variable]: A tuple with two elements. Each element is a Variable.
        The first one is k largest elements along each last
        dimensional slice. The second one is indices of values
F
fengjiayi 已提交
4831
        within the last dimension of input.
Q
qingqing01 已提交
4832

F
fengjiayi 已提交
4833 4834
    Raises:
        ValueError: If k < 1 or k is not less than the last dimension of input
Q
qingqing01 已提交
4835 4836 4837 4838 4839 4840 4841

    Examples:
        .. code-block:: python

            top5_values, top5_indices = layers.topk(input, k=5)
    """
    helper = LayerHelper("top_k", **locals())
X
Xin Pan 已提交
4842 4843
    values = helper.create_variable_for_type_inference(dtype=input.dtype)
    indices = helper.create_variable_for_type_inference(dtype="int64")
W
whs 已提交
4844 4845 4846 4847 4848 4849
    inputs = {"X": [input]}
    attrs = None
    if isinstance(k, Variable):
        inputs['K'] = k
    else:
        attrs = {'k': k}
Q
qingqing01 已提交
4850 4851
    helper.append_op(
        type="top_k",
W
whs 已提交
4852
        inputs=inputs,
Q
qingqing01 已提交
4853 4854
        outputs={"Out": [values],
                 "Indices": [indices]},
W
whs 已提交
4855
        attrs=attrs)
Q
qingqing01 已提交
4856 4857 4858 4859 4860
    values.stop_gradient = True
    indices.stop_gradient = True
    return values, indices


4861
def edit_distance(input, label, normalized=True, ignored_tokens=None):
4862
    """
Y
ying 已提交
4863 4864 4865 4866 4867 4868 4869 4870 4871
    EditDistance operator computes the edit distances between a batch of
    hypothesis strings and their references. Edit distance, also called
    Levenshtein distance, measures how dissimilar two strings are by counting
    the minimum number of operations to transform one string into anthor.
    Here the operations include insertion, deletion, and substitution.

    For example, given hypothesis string A = "kitten" and reference
    B = "sitting", the edit distance is 3 for A will be transformed into B
    at least after two substitutions and one insertion:
W
wanghaoshuang 已提交
4872

Y
ying 已提交
4873
    "kitten" -> "sitten" -> "sittin" -> "sitting"
W
wanghaoshuang 已提交
4874

4875
    The input is a LoDTensor consisting of all the hypothesis strings with
Y
ying 已提交
4876 4877
    the total number denoted by `batch_size`, and the separation is specified
    by the LoD information. And the `batch_size` reference strings are arranged
4878
    in order in the same way in the input LoDTensor.
W
wanghaoshuang 已提交
4879

4880
    The output contains the `batch_size` results and each stands for the edit
Y
ying 已提交
4881 4882
    distance for a pair of strings respectively. If Attr(normalized) is true,
    the edit distance will be divided by the length of reference string.
W
wanghaoshuang 已提交
4883

4884 4885 4886
    Args:
        input(Variable): The indices for hypothesis strings.
        label(Variable): The indices for reference strings.
4887
        normalized(bool, default True): Indicated whether to normalize the edit distance by
Y
ying 已提交
4888
                          the length of reference string.
4889
        ignored_tokens(list<int>, default None): Tokens that should be removed before
Y
ying 已提交
4890
                                     calculating edit distance.
4891
        name (str): The name of this layer. It is optional.
4892

W
wanghaoshuang 已提交
4893
    Returns:
W
wanghaoshuang 已提交
4894
        Variable: sequence-to-sequence edit distance in shape [batch_size, 1].
W
wanghaoshuang 已提交
4895 4896 4897 4898

    Examples:
        .. code-block:: python

T
tink2123 已提交
4899 4900
            x = fluid.layers.data(name='x', shape=[1], dtype='float32')
            y = fluid.layers.data(name='y', shape=[1], dtype='float32')
4901
            cost = fluid.layers.edit_distance(input=x,label=y)
4902
    """
4903
    helper = LayerHelper("edit_distance", **locals())
4904

4905
    # remove some tokens from input and labels
W
wanghaoshuang 已提交
4906
    if ignored_tokens is not None and len(ignored_tokens) > 0:
X
Xin Pan 已提交
4907 4908
        erased_input = helper.create_variable_for_type_inference(dtype="int64")
        erased_label = helper.create_variable_for_type_inference(dtype="int64")
4909 4910 4911 4912 4913

        helper.append_op(
            type="sequence_erase",
            inputs={"X": [input]},
            outputs={"Out": [erased_input]},
W
wanghaoshuang 已提交
4914
            attrs={"tokens": ignored_tokens})
4915 4916 4917 4918 4919
        input = erased_input

        helper.append_op(
            type="sequence_erase",
            inputs={"X": [label]},
W
whs 已提交
4920
            outputs={"Out": [erased_label]},
W
wanghaoshuang 已提交
4921
            attrs={"tokens": ignored_tokens})
4922 4923
        label = erased_label

4924
    # edit distance op
X
Xin Pan 已提交
4925 4926
    edit_distance_out = helper.create_variable_for_type_inference(dtype="int64")
    sequence_num = helper.create_variable_for_type_inference(dtype="int64")
4927 4928 4929 4930
    helper.append_op(
        type="edit_distance",
        inputs={"Hyps": [input],
                "Refs": [label]},
4931 4932
        outputs={"Out": [edit_distance_out],
                 "SequenceNum": [sequence_num]},
4933 4934
        attrs={"normalized": normalized})

4935
    return edit_distance_out, sequence_num
4936 4937 4938 4939 4940


def ctc_greedy_decoder(input, blank, name=None):
    """
    This op is used to decode sequences by greedy policy by below steps:
Y
yi.wu 已提交
4941

Y
ying 已提交
4942 4943 4944 4945
    1. Get the indexes of max value for each row in input. a.k.a.
       numpy.argmax(input, axis=0).
    2. For each sequence in result of step1, merge repeated tokens between two
       blanks and delete all blanks.
4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962

    A simple example as below:

    .. code-block:: text

        Given:

        input.data = [[0.6, 0.1, 0.3, 0.1],
                      [0.3, 0.2, 0.4, 0.1],
                      [0.1, 0.5, 0.1, 0.3],
                      [0.5, 0.1, 0.3, 0.1],

                      [0.5, 0.1, 0.3, 0.1],
                      [0.2, 0.2, 0.2, 0.4],
                      [0.2, 0.2, 0.1, 0.5],
                      [0.5, 0.1, 0.3, 0.1]]

4963
        input.lod = [[4, 4]]
M
minqiyang 已提交
4964

W
whs 已提交
4965
        Computation:
4966

W
whs 已提交
4967 4968 4969 4970 4971 4972
        step1: Apply argmax to first input sequence which is input.data[0:4]. Then we get:
               [[0], [2], [1], [0]]
        step2: merge repeated tokens and remove blank which is 0. Then we get first output sequence:
               [[2], [1]]

        Finally:
4973 4974 4975 4976 4977

        output.data = [[2],
                       [1],
                       [3]]

4978
        output.lod = [[2, 1]]
4979

W
whs 已提交
4980

4981 4982
    Args:

Y
ying 已提交
4983 4984 4985 4986 4987 4988 4989 4990 4991
        input(Variable): (LoDTensor<float>), the probabilities of
                         variable-length sequences, which is a 2-D Tensor with
                         LoD information. It's shape is [Lp, num_classes + 1],
                         where Lp is the sum of all input sequences' length and
                         num_classes is the true number of classes. (not
                         including the blank label).
        blank(int): the blank label index of Connectionist Temporal
                    Classification (CTC) loss, which is in thehalf-opened
                    interval [0, num_classes + 1).
4992
        name (str): The name of this layer. It is optional.
4993 4994

    Returns:
H
haowang101779990 已提交
4995 4996 4997
        Variable: CTC greedy decode result which is a 2-D tensor with shape [Lp, 1]. \
                  'Lp' is the sum if all output sequences' length. If all the sequences \
                  in result were empty, the result LoDTensor will be [-1] with  \
M
minqiyang 已提交
4998
                  LoD [[]] and dims [1, 1].
4999 5000 5001 5002 5003

    Examples:
        .. code-block:: python

            x = fluid.layers.data(name='x', shape=[8], dtype='float32')
W
wanghaoshuang 已提交
5004

5005
            cost = fluid.layers.ctc_greedy_decoder(input=x, blank=0)
W
wanghaoshuang 已提交
5006
    """
5007
    helper = LayerHelper("ctc_greedy_decoder", **locals())
Q
qingqing01 已提交
5008
    _, topk_indices = topk(input, k=1)
5009 5010

    # ctc align op
X
Xin Pan 已提交
5011
    ctc_out = helper.create_variable_for_type_inference(dtype="int64")
5012 5013 5014
    helper.append_op(
        type="ctc_align",
        inputs={"Input": [topk_indices]},
W
wanghaoshuang 已提交
5015
        outputs={"Output": [ctc_out]},
5016 5017
        attrs={"merge_repeated": True,
               "blank": blank})
5018
    return ctc_out
5019 5020


W
Wu Yi 已提交
5021
def warpctc(input, label, blank=0, norm_by_times=False, use_cudnn=False):
W
wanghaoshuang 已提交
5022
    """
5023 5024
    An operator integrating the open source Warp-CTC library
    (https://github.com/baidu-research/warp-ctc)
W
wanghaoshuang 已提交
5025
    to compute Connectionist Temporal Classification (CTC) loss.
5026 5027
    It can be aliased as softmax with CTC, since a native softmax activation is
    interated to the Warp-CTC library, to to normlize values for each row of the
W
wanghaoshuang 已提交
5028 5029 5030
    input tensor.

    Args:
5031
       input (Variable): The unscaled probabilities of variable-length sequences,
W
wanghaoshuang 已提交
5032 5033 5034 5035
         which is a 2-D Tensor with LoD information.
         It's shape is [Lp, num_classes + 1], where Lp is the sum of all input
         sequences' length and num_classes is the true number of classes.
         (not including the blank label).
5036
       label (Variable): The ground truth of variable-length sequence,
5037 5038 5039
         which is a 2-D Tensor with LoD information. It is of the shape [Lg, 1],
         where Lg is th sum of all labels' length.
       blank (int, default 0): The blank label index of Connectionist
W
wanghaoshuang 已提交
5040 5041
         Temporal Classification (CTC) loss, which is in the
         half-opened interval [0, num_classes + 1).
5042 5043 5044
       norm_by_times(bool, default false): Whether to normalize the gradients
         by the number of time-step, which is also the sequence's length.
         There is no need to normalize the gradients if warpctc layer was
5045
         follewed by a mean_op.
W
Wu Yi 已提交
5046
       use_cudnn (bool, default false): Whether to use cudnn.
W
wanghaoshuang 已提交
5047 5048

    Returns:
5049 5050
        Variable: The Connectionist Temporal Classification (CTC) loss,
        which is a 2-D Tensor of the shape [batch_size, 1].
W
wanghaoshuang 已提交
5051 5052

    Examples:
5053

W
wanghaoshuang 已提交
5054
        .. code-block:: python
5055

5056 5057 5058
            label = fluid.layers.data(shape=[11, 8], dtype='float32', lod_level=1)
            predict = fluid.layers.data(shape=[11, 1], dtype='float32')
            cost = fluid.layers.warpctc(input=predict, label=label)
W
wanghaoshuang 已提交
5059 5060

    """
F
fengjiayi 已提交
5061
    helper = LayerHelper('warpctc', **locals())
X
Xin Pan 已提交
5062 5063
    loss_out = helper.create_variable_for_type_inference(dtype=input.dtype)
    grad_out = helper.create_variable_for_type_inference(dtype=input.dtype)
W
wanghaoshuang 已提交
5064 5065 5066 5067 5068 5069
    helper.append_op(
        type='warpctc',
        inputs={'Logits': [input],
                'Label': [label]},
        outputs={'WarpCTCGrad': [grad_out],
                 'Loss': [loss_out]},
W
Wu Yi 已提交
5070 5071 5072 5073 5074
        attrs={
            'blank': blank,
            'norm_by_times': norm_by_times,
            'use_cudnn': use_cudnn
        })
W
wanghaoshuang 已提交
5075
    return loss_out
5076 5077 5078 5079 5080 5081 5082 5083 5084 5085 5086 5087 5088 5089 5090


def sequence_reshape(input, new_dim):
    """
    **Sequence Reshape Layer**

    This layer will rearrange the input sequences. The new dimension is set by
    user. Length of each sequence is computed according to original length,
    original dimension and new dimension. The following example will help to
    illustrate the function of this layer:

    .. code-block:: text

        x is a LoDTensor:
            x.lod  = [[0, 2, 6]]
5091 5092 5093
            x.data = [[1,  2], [3,  4],
                      [5,  6], [7,  8],
                      [9, 10], [11, 12]]
5094 5095 5096 5097 5098
            x.dims = [6, 2]

        set new_dim = 4

        then out is a LoDTensor:
5099

5100
            out.lod  = [[0, 1, 3]]
5101 5102 5103 5104

            out.data = [[1,  2,  3,  4],
                        [5,  6,  7,  8],
                        [9, 10, 11, 12]]
5105 5106 5107 5108 5109 5110 5111
            out.dims = [3, 4]

    Currently, only 1-level LoDTensor is supported and please make sure
    (original length * original dimension) can be divided by new dimension with
    no remainder for each sequence.

    Args:
5112 5113 5114

       input (Variable): A 2-D LoDTensor with shape being [N, M] where M for dimension.
       new_dim (int): New dimension that the input LoDTensor is reshaped to.
5115 5116

    Returns:
5117

5118 5119 5120 5121 5122
        Variable: Reshaped LoDTensor according to new dimension.

    Examples:
        .. code-block:: python

5123
            x = fluid.layers.data(shape=[5, 20], dtype='float32', lod_level=1)
5124
            x_reshaped = fluid.layers.sequence_reshape(input=x, new_dim=10)
5125 5126
    """
    helper = LayerHelper('sequence_reshape', **locals())
X
Xin Pan 已提交
5127
    out = helper.create_variable_for_type_inference(helper.input_dtype())
5128 5129 5130 5131 5132 5133
    helper.append_op(
        type='sequence_reshape',
        inputs={'X': [input]},
        outputs={'Out': [out]},
        attrs={'new_dim': new_dim})
    return out
Y
ying 已提交
5134 5135


5136 5137 5138 5139
# FIXME(wuyi): let docstring_checker.py understand @autodoc.
# For now, the comments in c++ use types like Tensor, but in python side
# the type is often "Variable", and arguments may vary.
@templatedoc(op_type="nce")
Y
Yang Yu 已提交
5140 5141 5142 5143 5144 5145
def nce(input,
        label,
        num_total_classes,
        sample_weight=None,
        param_attr=None,
        bias_attr=None,
C
chengduo 已提交
5146
        num_neg_samples=None,
5147 5148 5149
        name=None,
        sampler="uniform",
        custom_dist=None,
5150 5151
        seed=0,
        is_sparse=False):
5152 5153 5154 5155 5156 5157 5158
    """
    ${comment}

    Args:
        input (Variable): input variable.
        label (Variable): label.
        num_total_classes (int):${num_total_classes_comment}
5159 5160
        sample_weight (Variable|None): A Variable of shape [batch_size, 1]
            storing a weight for each sample. The default weight for each
5161
            sample is 1.0.
C
chengduo 已提交
5162 5163 5164 5165 5166 5167 5168 5169 5170
        param_attr (ParamAttr|None): The parameter attribute for learnable parameters/weights
             of nce. If it is set to None or one attribute of ParamAttr, nce
             will create ParamAttr as param_attr. If the Initializer of the param_attr
             is not set, the parameter is initialized with Xavier. Default: None.
        bias_attr (ParamAttr|bool|None): The parameter attribute for the bias of nce.
             If it is set to False, no bias will be added to the output units.
             If it is set to None or one attribute of ParamAttr, nce
             will create ParamAttr as bias_attr. If the Initializer of the bias_attr
             is not set, the bias is initialized zero. Default: None.
5171
        num_neg_samples (int): ${num_neg_samples_comment}
C
chengduo 已提交
5172 5173
        name (str|None): A name for this layer(optional). If set None, the layer
             will be named automatically. Default: None.
5174 5175 5176
        sampler (str): The sampler used to sample class from negtive classes.
                       It can be 'uniform', 'log_uniform' or 'custom_dist'.
                       default: 'uniform'.
5177
        custom_dist (float[]): A float[] with size=num_total_classes.
5178 5179 5180 5181
                       It is used when sampler is set to 'custom_dist'.
                       custom_dist[i] is the probsbility of i-th class to be sampled.
                       default: None.
        seed (int): The seed used in sampler. default: 0.
5182
        is_sparse(bool): The flag indicating whether to use sparse update, the weight@GRAD and bias@GRAD will be changed to SelectedRows.
F
fengjiayi 已提交
5183

5184
    Returns:
Y
Yibing Liu 已提交
5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200 5201 5202 5203 5204 5205 5206 5207 5208 5209 5210 5211
        Variable: The output nce loss.

    Examples:
        .. code-block:: python

            window_size = 5
            words = []
            for i in xrange(window_size):
                words.append(layers.data(
                    name='word_{0}'.format(i), shape=[1], dtype='int64'))

            dict_size = 10000
            label_word = int(window_size / 2) + 1

            embs = []
            for i in xrange(window_size):
                if i == label_word:
                    continue

                emb = layers.embedding(input=words[i], size=[dict_size, 32],
                                       param_attr='emb.w', is_sparse=True)
                embs.append(emb)

            embs = layers.concat(input=embs, axis=1)
            loss = layers.nce(input=embs, label=words[label_word],
                          num_total_classes=dict_size, param_attr='nce.w',
                          bias_attr='nce.b')
5212 5213 5214 5215 5216 5217 5218 5219 5220

            #or use custom distribution
            dist = fluid.layers.assign(input=np.array([0.05,0.5,0.1,0.3,0.05]).astype("float32"))
            loss = layers.nce(input=embs, label=words[label_word],
                          num_total_classes=5, param_attr='nce.w',
                          bias_attr='nce.b',
                          num_neg_samples=3,
                          sampler="custom_dist",
                          custom_dist=dist)
5221

5222
    """
Y
Yang Yu 已提交
5223 5224 5225
    helper = LayerHelper('nce', **locals())
    assert isinstance(input, Variable)
    assert isinstance(label, Variable)
C
chengduo 已提交
5226 5227

    dim = input.shape[1]
Y
Yang Yu 已提交
5228 5229 5230 5231 5232 5233
    num_true_class = label.shape[1]
    w = helper.create_parameter(
        attr=helper.param_attr,
        shape=[num_total_classes, dim],
        is_bias=False,
        dtype=input.dtype)
5234
    inputs = {}
C
chengduo 已提交
5235 5236 5237 5238 5239 5240 5241
    if helper.bias_attr:
        b = helper.create_parameter(
            attr=helper.bias_attr,
            shape=[num_total_classes, 1],
            is_bias=True,
            dtype=input.dtype)
        inputs['Bias'] = b
X
Xin Pan 已提交
5242 5243 5244
    cost = helper.create_variable_for_type_inference(dtype=input.dtype)
    sample_logits = helper.create_variable_for_type_inference(dtype=input.dtype)
    sample_labels = helper.create_variable_for_type_inference(dtype=label.dtype)
Y
Yang Yu 已提交
5245

5246 5247 5248 5249
    inputs['Input'] = input
    inputs['Label'] = label
    inputs['Weight'] = w
    inputs['SampleWeight'] = sample_weight if sample_weight is not None else []
5250 5251 5252 5253 5254 5255 5256

    if sampler == "uniform":
        sampler = 0
    elif sampler == "log_uniform":
        sampler = 1
    elif sampler == "custom_dist":
        assert custom_dist is not None
5257 5258 5259 5260 5261 5262 5263 5264 5265
        # assert isinstance(custom_dist, Variable)

        custom_dist_len = len(custom_dist)
        alias_probs_ = [0] * custom_dist_len
        alias_ = [0] * custom_dist_len
        bigs = []
        littles = []
        for i in range(custom_dist_len):
            normal_prob = custom_dist[i] * custom_dist_len
5266
            if normal_prob - 1.0 > 0:
5267
                bigs.append((i, normal_prob))
5268
            elif 1.0 - normal_prob > 0:
5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283
                littles.append((i, normal_prob))
            else:
                alias_probs_[i] = normal_prob
                alias_[i] = -1

        while len(bigs) and len(littles):
            big = bigs.pop(0)
            little = littles.pop(0)

            big_idx = big[0]
            big_prob = big[1]

            alias_probs_[little[0]] = little[1]
            alias_[little[0]] = big_idx
            big_left = big[1] + little[1] - 1
5284
            if big_left - 1.0 > 0:
5285
                bigs.append((big_idx, big_left))
5286
            elif 1.0 - big_left > 0:
5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298 5299 5300
                littles.append((big_idx, big_left))
            else:
                alias_probs_[big_idx] = big_left
                alias_[big_idx] = -1

        if len(bigs):
            big = bigs.pop(0)
            alias_probs_[big[0]] = 1.0
            alias_[big[0]] = -1
        if len(littles):
            little = littles.pop(0)
            alias_probs_[little[0]] = 1.0
            alias_[little[0]] = -1

5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312 5313 5314 5315
        def _init_by_numpy_array(numpy_array):
            ret = helper.create_parameter(
                attr=ParamAttr(),
                shape=numpy_array.shape,
                dtype=numpy_array.dtype,
                default_initializer=NumpyArrayInitializer(numpy_array))
            ret.stop_gradient = True
            return ret

        inputs['CustomDistProbs'] = _init_by_numpy_array(
            np.array(custom_dist).astype('float32'))
        inputs['CustomDistAlias'] = _init_by_numpy_array(
            np.array(alias_).astype('int32'))
        inputs['CustomDistAliasProbs'] = _init_by_numpy_array(
            np.array(alias_probs_).astype('float32'))
5316 5317 5318 5319
        sampler = 2
    else:
        raise Exception("Unsupported sampler type.")

5320 5321 5322 5323 5324
    if num_neg_samples is None:
        num_neg_samples = 10
    else:
        num_neg_samples = int(num_neg_samples)

5325 5326 5327 5328
    remote_prefetch = is_sparse
    print(
        "With sparse mode, if your models has only small parameter prefetch may cause speed down"
    )
5329

Y
Yang Yu 已提交
5330 5331
    attrs = {
        'num_total_classes': int(num_total_classes),
5332 5333
        'num_neg_samples': num_neg_samples,
        'seed': seed,
5334
        'sampler': sampler,
5335 5336
        'is_sparse': is_sparse,
        'remote_prefetch': remote_prefetch
Y
Yang Yu 已提交
5337
    }
Y
Yang Yu 已提交
5338 5339 5340

    helper.append_op(
        type='nce',
C
chengduo 已提交
5341
        inputs=inputs,
Y
Yang Yu 已提交
5342 5343 5344 5345 5346 5347
        outputs={
            'Cost': cost,
            'SampleLogits': sample_logits,
            'SampleLabels': sample_labels
        },
        attrs=attrs)
Y
Yang Yu 已提交
5348
    return cost / (num_neg_samples + 1)
5349 5350


C
chengduo 已提交
5351 5352
def hsigmoid(input,
             label,
5353
             num_classes,
C
chengduo 已提交
5354 5355
             param_attr=None,
             bias_attr=None,
J
JiabinYang 已提交
5356
             name=None,
5357 5358 5359
             path_table=None,
             path_code=None,
             is_custom=False,
J
JiabinYang 已提交
5360
             is_sparse=False):
W
weixing02 已提交
5361 5362
    """
    The hierarchical sigmoid operator is used to accelerate the training
M
minqiyang 已提交
5363
    process of language model. This operator organizes the classes into a
M
minqiyang 已提交
5364
    complete binary tree, or you can use is_custom to pass your own tree to
5365
    implement hierarchical. Each leaf node represents a class(a word) and each
G
guosheng 已提交
5366 5367 5368 5369 5370 5371
    internal node acts as a binary classifier. For each word there's a unique
    path from root to it's leaf node, hsigmoid calculate the cost for each
    internal node on the path, and sum them to get a total cost. hsigmoid can
    achive a acceleration from :math:`O(N)` to :math:`O(logN)`, where :math:`N`
    represents the size of word dict.

5372
    Using default tree you can Refer to `Hierarchical Probabilistic Neural Network Language Model
G
guosheng 已提交
5373
    <http://www.iro.umontreal.ca/~lisa/pointeurs/hierarchical-nnlm-aistats05.pdf>`_
M
minqiyang 已提交
5374

5375 5376
    And if you want to use the costumed tree by set 'is_custom' as true you may need to do following things first:

H
haowang101779990 已提交
5377 5378 5379 5380
    1. using your word dict to build a binary tree, each leaf node should be an word of your word dict
    2. build a dict to store word_id -> word's leaf to root path, we call it path_table.
    3. build a dict to store word_id -> code of word's leaf to root path, we call it path_code. Code
       means label of each binary classification, using 1 indicate true, 0 indicate false.
M
minqiyang 已提交
5381
    4. now, each word should has its path and code along the path, you can pass a batch of path and code
H
haowang101779990 已提交
5382
       related to the same batch of inputs.
5383

W
weixing02 已提交
5384
    Args:
M
minqiyang 已提交
5385
        input (Variable): The input tensor variable with shape
G
guosheng 已提交
5386 5387 5388 5389
            :math:`[N \\times D]`, where :math:`N` is the size of mini-batch,
            and :math:`D` is the feature size.
        label (Variable): The tensor variable contains labels of training data.
            It's a tensor with shape is :math:`[N \\times 1]`.
M
minqiyang 已提交
5390 5391
        num_classes: (int), The number of classes, must not be less than 2. with default tree this has to be set,
            it should never be None under is_custom=False, but while is_custom is true, it should be non leaf num
5392
            which indicates the num of classes using by binary classify.
C
chengduo 已提交
5393 5394 5395 5396 5397 5398 5399 5400 5401 5402 5403
        param_attr (ParamAttr|None): The parameter attribute for learnable parameters/weights
             of hsigmoid. If it is set to None or one attribute of ParamAttr, hsigmoid
             will create ParamAttr as param_attr. If the Initializer of the param_attr
             is not set, the parameter is initialized with Xavier. Default: None.
        bias_attr (ParamAttr|bool|None): The parameter attribute for the bias of hsigmoid.
             If it is set to False, no bias will be added to the output units.
             If it is set to None or one attribute of ParamAttr, hsigmoid
             will create ParamAttr as bias_attr. If the Initializer of the bias_attr
             is not set, the bias is initialized zero. Default: None.
        name (str|None): A name for this layer(optional). If set None, the layer
             will be named automatically. Default: None.
M
minqiyang 已提交
5404
        path_table: (Variable|None) this variable can store each batch of samples' path to root,
5405
            it should be in leaf -> root order
M
minqiyang 已提交
5406 5407 5408
            path_table should have the same shape with path_code, and for each sample i path_table[i] indicates a np.array like
            structure and each element in this array is indexes in parent nodes' Weight Matrix.
        path_code:  (Variable|None) this variable can store each batch of samples' code,
5409
            each code consist with every code of parent nodes. it should be in leaf -> root order
M
minqiyang 已提交
5410
        is_custom: (bool|False)using user defined binary tree instead of default complete binary tree, if costum is
5411
             set you need to set path_table/path_code/num_classes, otherwise num_classes should be set
M
minqiyang 已提交
5412
        is_sparse: (bool|False)using sparse update instead of dense update, if set, the gradient
5413
             of W and input will be sparse.
W
weixing02 已提交
5414 5415

    Returns:
J
JiabinYang 已提交
5416
        Out: (LodTensor) The cost of hierarchical sigmoid operator. the shape is [N, 1]
W
weixing02 已提交
5417 5418 5419 5420 5421

    Examples:

        .. code-block:: python

G
guosheng 已提交
5422 5423 5424
            x = fluid.layers.data(name='x', shape=[2], dtype='float32')
            y = fluid.layers.data(name='y', shape=[1], dtype='int64')
            out = fluid.layers.hsigmoid(input=x, label=y, num_classes=6)
W
weixing02 已提交
5425 5426 5427 5428
    """

    helper = LayerHelper('hierarchical_sigmoid', **locals())
    dtype = helper.input_dtype()
X
Xin Pan 已提交
5429 5430
    out = helper.create_variable_for_type_inference(dtype)
    pre_out = helper.create_variable_for_type_inference(dtype)
W
weixing02 已提交
5431
    dim = input.shape[1]
5432
    if ((num_classes is None) or (num_classes < 2)) and (not is_custom):
J
JiabinYang 已提交
5433 5434 5435
        raise ValueError(
            "num_classes must not be less than 2 with default tree")

5436 5437 5438 5439
    if (is_custom) and (path_code is None):
        raise ValueError("path_code should not be None with costum tree")
    elif (is_custom) and (path_table is None):
        raise ValueError("path_table should not be None with costum tree")
5440 5441
    elif (is_custom) and (num_classes is None):
        raise ValueError("num_classes should not be None with costum tree")
5442 5443 5444
    else:
        pass

J
JiabinYang 已提交
5445
    weights = None
5446 5447 5448 5449
    remote_prefetch = is_sparse
    print(
        "With sparse mode, if your models has only small parameter prefetch may cause speed down"
    )
5450
    if not is_custom:
J
JiabinYang 已提交
5451 5452 5453 5454 5455 5456 5457 5458
        weights = helper.create_parameter(
            attr=helper.param_attr,
            shape=[num_classes - 1, dim],
            is_bias=False,
            dtype=input.dtype)
    else:
        weights = helper.create_parameter(
            attr=helper.param_attr,
5459
            shape=[num_classes, dim],
J
JiabinYang 已提交
5460 5461
            is_bias=False,
            dtype=input.dtype)
5462 5463 5464
    inputs = {
        "X": input,
        "W": weights,
5465
        "PathTable": path_table,
5466
        "PathCode": path_code,
5467 5468
        "Label": label
    }
W
weixing02 已提交
5469
    if helper.bias_attr:
5470
        if not is_custom:
J
JiabinYang 已提交
5471 5472
            bias = helper.create_parameter(
                attr=helper.bias_attr,
J
JiabinYang 已提交
5473
                shape=[num_classes - 1, 1],
J
JiabinYang 已提交
5474 5475 5476 5477 5478 5479
                is_bias=True,
                dtype=input.dtype)
            inputs['Bias'] = bias
        else:
            bias = helper.create_parameter(
                attr=helper.bias_attr,
5480
                shape=[num_classes, 1],
J
JiabinYang 已提交
5481 5482 5483
                is_bias=True,
                dtype=input.dtype)
            inputs['Bias'] = bias
W
weixing02 已提交
5484 5485
    helper.append_op(
        type="hierarchical_sigmoid",
W
weixing02 已提交
5486
        inputs=inputs,
W
weixing02 已提交
5487
        outputs={"Out": out,
5488 5489 5490 5491 5492 5493 5494
                 "PreOut": pre_out,
                 "W_Out": weights},
        attrs={
            "num_classes": num_classes,
            "is_sparse": is_sparse,
            "remote_prefetch": remote_prefetch
        })
W
weixing02 已提交
5495 5496 5497
    return out


Y
fix ci.  
ying 已提交
5498
def transpose(x, perm, name=None):
Y
ying 已提交
5499 5500 5501 5502 5503 5504 5505
    """
    Permute the dimensions of `input` according to `perm`.

    The `i`-th dimension  of the returned tensor will correspond to the
    perm[i]-th dimension of `input`.

    Args:
5506 5507 5508
        x (Variable): The input Tensor.
        perm (list): A permutation of the dimensions of `input`.
        name (str): The name of this layer. It is optional.
Y
ying 已提交
5509 5510 5511 5512 5513 5514 5515

    Returns:
        Variable: A transposed Tensor.

    Examples:
        .. code-block:: python

5516
            # use append_batch_size=False to avoid prepending extra
5517
            # batch size in shape
5518
            x = fluid.layers.data(name='x', shape=[5, 10, 15],
5519
                            dtype='float32', append_batch_size=False)
Y
fix ci.  
ying 已提交
5520
            x_transposed = layers.transpose(x, perm=[1, 0, 2])
Y
ying 已提交
5521 5522
    """

Y
fix ci.  
ying 已提交
5523
    if len(perm) != len(x.shape):
Y
ying 已提交
5524 5525 5526
        raise ValueError(
            "Input(perm) is the permutation of dimensions of Input(input). "
            "It's length shoud be equal to Input(input)'s rank.")
Y
ying 已提交
5527 5528 5529 5530 5531 5532
    for idx, dim in enumerate(perm):
        if dim >= len(x.shape):
            raise ValueError(
                "Each element in perm should be less than x's rank. "
                "%d-th element in perm is %d which accesses x's rank %d." %
                (idx, perm[idx], len(x.shape)))
Y
ying 已提交
5533 5534

    helper = LayerHelper('transpose', **locals())
X
Xin Pan 已提交
5535 5536
    out = helper.create_variable_for_type_inference(x.dtype)
    x_shape = helper.create_variable_for_type_inference(x.dtype)
Y
ying 已提交
5537
    helper.append_op(
5538
        type='transpose2',
Y
fix ci.  
ying 已提交
5539
        inputs={'X': [x]},
5540 5541
        outputs={'Out': [out],
                 'XShape': [x_shape]},
Y
ying 已提交
5542 5543
        attrs={'axis': perm})
    return out
5544 5545


5546 5547 5548 5549 5550 5551 5552
def im2sequence(input,
                filter_size=1,
                stride=1,
                padding=0,
                input_image_size=None,
                out_stride=1,
                name=None):
5553
    """
5554 5555 5556 5557 5558 5559 5560
    Extracts image patches from the input tensor to form a tensor of shape
    {input.batch_size * output_height * output_width, filter_size_H *
    filter_size_W * input.channels} which is similar with im2col.
    This op use filter / kernel to scan images and convert these images to
    sequences. After expanding, the number of time step are
    output_height * output_width for an image, in which output_height and
    output_width are calculated by below equation:
5561 5562 5563 5564 5565 5566 5567 5568 5569 5570

    .. math::

        output\_size = 1 + \
            (2 * padding + img\_size - block\_size + stride - 1) / stride

    And the dimension of each time step is block_y * block_x * input.channels.

    Args:
        input (Variable): The input should be a tensor in NCHW format.
W
wanghaoshuang 已提交
5571 5572 5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583 5584 5585 5586 5587 5588

        filter_size(int|tuple|None): The filter size. If filter_size is a tuple,
            it must contain two integers, (filter_size_H, filter_size_W).
            Otherwise, the filter will be a square.

        stride(int|tuple): The stride size. If stride is a tuple, it must
            contain two integers, (stride_H, stride_W). Otherwise, the
            stride_H = stride_W = stride. Default: stride = 1.

        padding(int|tuple): The padding size. If padding is a tuple, it can
            contain two integers like (padding_H, padding_W) which means
            padding_up = padding_down = padding_H and
            padding_left = padding_right = padding_W. Or it can use
            (padding_up, padding_left, padding_down, padding_right) to indicate
            paddings of four direction. Otherwise, a scalar padding means
            padding_up = padding_down = padding_left = padding_right = padding
            Default: padding = 0.

5589 5590 5591 5592 5593 5594 5595 5596 5597
        input_image_size(Variable): the input contains image real size.It's dim
            is [batchsize, 2]. It is dispensable.It is just for batch inference.

        out_stride(int|tuple): The scaling of image through CNN. It is
            dispensable. It is valid only when input_image_size is not null.
            If out_stride is tuple,  it must contain two intergers,
            (out_stride_H, out_stride_W). Otherwise,
            the out_stride_H = out_stride_W = out_stride.

5598 5599 5600
        name (int): The name of this layer. It is optional.

    Returns:
W
wanghaoshuang 已提交
5601 5602 5603 5604 5605
        output: The output is a LoDTensor with shape
        {input.batch_size * output_height * output_width,
        filter_size_H * filter_size_W * input.channels}.
        If we regard output as a matrix, each row of this matrix is
        a step of a sequence.
5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616 5617 5618 5619 5620 5621 5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632

    Examples:

        .. code-block:: text

            Given:

            x = [[[[ 6.  2.  1.]
                   [ 8.  3.  5.]
                   [ 0.  2.  6.]]

                  [[ 2.  4.  4.]
                   [ 6.  3.  0.]
                   [ 6.  4.  7.]]]

                 [[[ 6.  7.  1.]
                   [ 5.  7.  9.]
                   [ 2.  4.  8.]]

                  [[ 1.  2.  1.]
                   [ 1.  3.  5.]
                   [ 9.  0.  8.]]]]

            x.dims = {2, 2, 3, 3}

            And:

W
wanghaoshuang 已提交
5633 5634 5635
            filter = [2, 2]
            stride = [1, 1]
            padding = [0, 0]
5636 5637 5638 5639 5640 5641 5642 5643 5644 5645 5646 5647

            Then:

            output.data = [[ 6.  2.  8.  3.  2.  4.  6.  3.]
                           [ 2.  1.  3.  5.  4.  4.  3.  0.]
                           [ 8.  3.  0.  2.  6.  3.  6.  4.]
                           [ 3.  5.  2.  6.  3.  0.  4.  7.]
                           [ 6.  7.  5.  7.  1.  2.  1.  3.]
                           [ 7.  1.  7.  9.  2.  1.  3.  5.]
                           [ 5.  7.  2.  4.  1.  3.  9.  0.]
                           [ 7.  9.  4.  8.  3.  5.  0.  8.]]

5648
            output.dims = {8, 8}
5649

5650
            output.lod = [[4, 4]]
5651

T
Tink_Y 已提交
5652
    Examples:
5653 5654 5655

        .. code-block:: python

5656 5657
            output = fluid.layers.im2sequence(
                input=layer, stride=[1, 1], filter_size=[2, 2])
5658 5659

    """
W
wanghaoshuang 已提交
5660 5661 5662 5663 5664 5665 5666 5667 5668 5669

    if isinstance(filter_size, int):
        filter_size = [filter_size, filter_size]
    if isinstance(stride, int):
        stride = [stride, stride]
    if isinstance(padding, int):
        padding = [padding, padding]
    if len(padding) == 2:
        padding.append(padding[0])
        padding.append(padding[1])
5670 5671 5672 5673 5674 5675 5676
    inputs = {"X": input}
    attrs = {"kernels": filter_size, "strides": stride, "padding": padding}
    if input_image_size:
        if isinstance(out_stride, int):
            out_stride = [out_stride, out_stride]
        inputs["Y"] = input_image_size
        attrs["out_stride"] = out_stride
5677
    helper = LayerHelper('im2sequence', **locals())
X
Xin Pan 已提交
5678
    out = helper.create_variable_for_type_inference(dtype=helper.input_dtype())
5679
    helper.append_op(
5680
        type='im2sequence', inputs=inputs, outputs={'Out': out}, attrs=attrs)
5681
    return out
5682 5683


Y
yuyang18 已提交
5684
@templatedoc()
5685
def row_conv(input, future_context_size, param_attr=None, act=None):
Y
yuyang18 已提交
5686 5687
    """
    ${comment}
5688 5689

    Args:
Y
yuyang18 已提交
5690
        input (${x_type}): ${x_comment}.
Y
yangyaming 已提交
5691 5692
        future_context_size (int): Future context size. Please note, the shape
            of convolution kernel is [future_context_size + 1, D].
5693 5694 5695 5696 5697
        param_attr (ParamAttr): Attributes of parameters, including
            name, initializer etc.
        act (str): Non-linear activation to be applied to output variable.

    Returns:
Y
yuyang18 已提交
5698
        ${out_comment}.
5699 5700

    Examples:
Y
yuyang18 已提交
5701 5702 5703 5704
        >>> import paddle.fluid as fluid
        >>> x = fluid.layers.data(name='x', shape=[16],
        >>>                        dtype='float32', lod_level=1)
        >>> out = fluid.layers.row_conv(input=x, future_context_size=2)
5705 5706 5707 5708 5709 5710
    """
    helper = LayerHelper('row_conv', **locals())
    dtype = helper.input_dtype()
    filter_shape = [future_context_size + 1, input.shape[1]]
    filter_param = helper.create_parameter(
        attr=helper.param_attr, shape=filter_shape, dtype=dtype)
X
Xin Pan 已提交
5711
    out = helper.create_variable_for_type_inference(dtype)
5712 5713 5714 5715 5716
    helper.append_op(
        type='row_conv',
        inputs={'X': [input],
                'Filter': [filter_param]},
        outputs={'Out': [out]})
Y
yangyaming 已提交
5717
    return helper.append_activation(out)
5718 5719


Y
yuyang18 已提交
5720
@templatedoc()
5721 5722
def multiplex(inputs, index):
    """
Y
yuyang18 已提交
5723 5724 5725 5726 5727 5728 5729
    ${comment}

    >>> import paddle.fluid as fluid
    >>> x1 = fluid.layers.data(name='x1', shape=[4], dtype='float32')
    >>> x2 = fluid.layers.data(name='x2', shape=[4], dtype='float32')
    >>> index = fluid.layers.data(name='index', shape=[1], dtype='int32')
    >>> out = fluid.layers.multiplex(inputs=[x1, x2], index=index)
5730 5731

    Args:
Y
yuyang18 已提交
5732 5733
       inputs (list): ${x_comment}.
       index (${ids_type}): ${ids_comment}.
5734 5735

    Returns:
Y
yuyang18 已提交
5736
        ${out_comment}.
5737 5738
    """
    helper = LayerHelper('multiplex', **locals())
Y
yangyaming 已提交
5739 5740 5741 5742 5743

    if not isinstance(inputs, list) and len(inputs) < 2:
        raise ValueError("inputs should be a list object and contains at least "
                         "2 elements.")

X
Xin Pan 已提交
5744
    out = helper.create_variable_for_type_inference(inputs[0].dtype)
5745 5746 5747 5748 5749 5750
    helper.append_op(
        type='multiplex',
        inputs={'X': inputs,
                'Ids': index},
        outputs={'Out': [out]})
    return out
5751 5752


5753 5754 5755
def softmax_with_cross_entropy(logits,
                               label,
                               soft_label=False,
J
jerrywgz 已提交
5756
                               ignore_index=kIgnoreIndex,
5757
                               numeric_stable_mode=True,
5758
                               return_softmax=False):
5759 5760
    """
    **Softmax With Cross Entropy Operator.**
5761

5762 5763 5764 5765
    Cross entropy loss with softmax is used as the output layer extensively. This
    operator computes the softmax normalized values for each row of the input
    tensor, after which cross-entropy loss is computed. This provides a more
    numerically stable gradient.
5766

5767 5768 5769
    Because this operator performs a softmax on logits internally, it expects
    unscaled logits. This operator should not be used with the output of
    softmax operator since that would produce incorrect results.
5770

5771 5772 5773
    When the attribute soft_label is set false, this operators expects mutually
    exclusive hard labels, each sample in a batch is in exactly one class with a
    probability of 1.0. Each sample in the batch will have a single label.
5774

5775
    The equation is as follows:
5776

5777
    1) Hard label (one-hot label, so every sample has exactly one class)
5778

5779 5780 5781 5782
    .. math::

        loss_j =  -\\text{logit}_{label_j} +
        \\log\\left(\\sum_{i=0}^{K}\\exp(\\text{logit}_i)\\right), j = 1,..., K
5783

5784 5785 5786
    2) Soft label (each sample can have a distribution over all classes)

    .. math::
5787

5788 5789 5790 5791
        loss_j =  -\\sum_{i=0}^{K}\\text{label}_i
        \\left(\\text{logit}_i - \\log\\left(\\sum_{i=0}^{K}
        \\exp(\\text{logit}_i)\\right)\\right), j = 1,...,K

S
sneaxiy 已提交
5792 5793 5794
    3) If numeric_stable_mode is True, softmax is calculated first by:

    .. math::
5795

H
haowang101779990 已提交
5796
        max_j &= \\max_{i=0}^{K}{\\text{logit}_i}
S
sneaxiy 已提交
5797

H
haowang101779990 已提交
5798
        log\\_max\\_sum_j &= \\log\\sum_{i=0}^{K}\\exp(logit_i - max_j)
S
sneaxiy 已提交
5799

H
haowang101779990 已提交
5800
        softmax_j &= \\exp(logit_j - max_j - {log\\_max\\_sum}_j)
S
sneaxiy 已提交
5801 5802 5803

    and then cross entropy loss is calculated by softmax and label.

5804 5805 5806 5807 5808 5809 5810 5811
    Args:
        logits (Variable): The unscaled log probabilities, which is a 2-D tensor
            with shape [N x K]. N is the batch_size, and K is the class number.
        label (Variable): The ground truth which is a 2-D tensor. If soft_label
            is set to false, Label is a Tensor<int64> with shape [N x 1]. If
            soft_label is set to true, Label is a Tensor<float/double> with
        soft_label (bool): A flag to indicate whether to interpretate the given
            labels as soft labels. By default, `soft_label` is set to False.
M
minqiyang 已提交
5812 5813
        ignore_index (int): Specifies a target value that is ignored and does
                            not contribute to the input gradient. Only valid
J
jerrywgz 已提交
5814
                            if soft_label is set to False. Default: kIgnoreIndex
S
sneaxiy 已提交
5815 5816 5817
        numeric_stable_mode (bool): A flag to indicate whether to use a more
                                    numerically stable algorithm. Only valid
                                    when soft_label is False and GPU is used.
5818 5819 5820
                                    When soft_label is True or CPU is used,
                                    the algorithm is always numerically stable.
                                    Note that the speed may be slower when use
5821
                                    stable algorithm. Default: True
5822
        return_softmax (bool): A flag indicating whether to return the softmax
5823
                               along with the cross entropy loss. Default: False
5824

5825
    Returns:
H
haowang101779990 已提交
5826 5827 5828 5829 5830
        Variable or Tuple of two Variables: Return the cross entropy loss if \
                                            `return_softmax` is False, otherwise the tuple \
                                            (loss, softmax), where the cross entropy loss is \
                                            a 2-D tensor with shape [N x 1], and softmax is a \
                                            2-D tensor with shape [N x K].
5831 5832 5833 5834 5835 5836 5837

    Examples:
        .. code-block:: python

            data = fluid.layers.data(name='data', shape=[128], dtype='float32')
            label = fluid.layers.data(name='label', shape=[1], dtype='int64')
            fc = fluid.layers.fc(input=data, size=100)
F
stash  
fengjiayi 已提交
5838 5839
            out = fluid.layers.softmax_with_cross_entropy(
                logits=fc, label=label)
5840 5841
    """
    helper = LayerHelper('softmax_with_cross_entropy', **locals())
X
Xin Pan 已提交
5842 5843
    softmax = helper.create_variable_for_type_inference(dtype=logits.dtype)
    loss = helper.create_variable_for_type_inference(dtype=logits.dtype)
5844 5845 5846 5847 5848 5849
    helper.append_op(
        type='softmax_with_cross_entropy',
        inputs={'Logits': logits,
                'Label': label},
        outputs={'Softmax': softmax,
                 'Loss': loss},
S
sneaxiy 已提交
5850 5851 5852 5853 5854
        attrs={
            'soft_label': soft_label,
            'ignore_index': ignore_index,
            'numeric_stable_mode': numeric_stable_mode
        })
5855 5856 5857 5858

    if return_softmax:
        return loss, softmax

5859 5860 5861
    return loss


5862 5863 5864
def sampled_softmax_with_cross_entropy(logits,
                                       label,
                                       num_samples,
X
xuezhong 已提交
5865
                                       num_true=1,
5866
                                       remove_accidental_hits=True,
X
xuezhong 已提交
5867 5868 5869
                                       use_customized_samples=False,
                                       customized_samples=None,
                                       customized_probabilities=None,
5870
                                       seed=0):
X
xuezhong 已提交
5871 5872 5873 5874 5875
    """
    **Sampled Softmax With Cross Entropy Operator.**

    Cross entropy loss with sampled softmax is used as the output layer for 
    larger output classes extensively. This operator samples a number of samples
5876
    for all examples, and computes the softmax normalized values for each 
X
xuezhong 已提交
5877 5878 5879 5880 5881 5882 5883 5884
    row of the sampled tensor, after which cross-entropy loss is computed. 

    Because this operator performs a softmax on logits internally, it expects
    unscaled logits. This operator should not be used with the output of
    softmax operator since that would produce incorrect results.
    
    For examples with T true labels (T >= 1), we assume that each true label has
    a probability of 1/T. For each sample, S samples are generated using a
X
xuezhong 已提交
5885
    log uniform distribution. True labels are concatenated with these samples to
X
xuezhong 已提交
5886 5887 5888 5889 5890 5891 5892 5893
    form T + S samples for each example. So, assume the shape of logits is
    [N x K], the shape for samples is [N x (T+S)]. For each sampled label, a 
    probability is calculated, which corresponds to the Q(y|x) in 
    [Jean et al., 2014](http://arxiv.org/abs/1412.2007).
    
    Logits are sampled according to the sampled labels. Then if 
    remove_accidental_hits is True, if a sample[i, j] accidentally hits true 
    labels, then the corresponding sampled_logits[i, j] is minus by 1e20 to 
X
xuezhong 已提交
5894
    make its softmax result close to zero. Then sampled logits are subtracted by
X
xuezhong 已提交
5895 5896 5897 5898 5899 5900 5901 5902 5903 5904 5905
    logQ(y|x), these sampled logits and re-indexed labels are used to compute 
    a softmax with cross entropy.

    Args:
        logits (Variable): The unscaled log probabilities, which is a 2-D tensor
            with shape [N x K]. N is the batch_size, and K is the class number.
        label (Variable): The ground truth which is a 2-D tensor. Label is a 
            Tensor<int64> with shape [N x T], where T is the number of true 
            labels per example. 
        num_samples (int): The number for each example, num_samples should be 
            less than the number of class.
5906
        num_true(int): The number of target classes per training example.
X
xuezhong 已提交
5907 5908 5909 5910 5911
        remove_accidental_hits (bool): A flag indicating whether to remove 
            accidental hits when sampling. If True and if a sample[i, j] 
            accidentally hits true labels, then the corresponding 
            sampled_logits[i, j] is minus by 1e20 to make its softmax result 
            close to zero. Default is True.
X
xuezhong 已提交
5912
        use_customized_samples (bool): Whether to use custom samples and probabities to sample
5913
            logits.
X
xuezhong 已提交
5914 5915 5916 5917 5918
        customized_samples (Variable): User defined samples, which is a 2-D tensor
            with shape [N, T + S]. S is the num_samples, and T is the number of true 
            labels per example. 
        customized_probabilities (Variable): User defined probabilities of samples, 
            a 2-D tensor which has the same shape with customized_samples.
5919 5920 5921
        seed (int): The random seed for generating random number, which is used
            in the process of sampling. Default is 0.

X
xuezhong 已提交
5922 5923 5924 5925 5926 5927 5928 5929 5930 5931 5932 5933 5934 5935 5936 5937 5938 5939 5940 5941
    Returns:
        Variable: Return the cross entropy loss which is a 2-D tensor with shape
                  [N x 1].

    Examples:
        .. code-block:: python

            logits = fluid.layers.data(name='data', shape=[256], dtype='float32')
            label = fluid.layers.data(name='label', shape=[5], dtype='int64')
            fc = fluid.layers.fc(input=data, size=100)
            out = fluid.layers.sampled_softmax_with_cross_entropy(
                logits=fc, label=label, num_samples=25)
    """
    helper = LayerHelper('sample_logits', **locals())
    samples = helper.create_variable_for_type_inference(dtype='int64')
    probabilities = helper.create_variable_for_type_inference(
        dtype=logits.dtype)
    sampled_logits \
        = helper.create_variable_for_type_inference(dtype=logits.dtype)
    sampled_label = helper.create_variable_for_type_inference(dtype='int64')
X
xuezhong 已提交
5942 5943
    sampled_softlabel = helper.create_variable_for_type_inference(
        dtype=logits.dtype)
X
xuezhong 已提交
5944 5945 5946 5947 5948

    helper.append_op(
        type='sample_logits',
        inputs={
            'Logits': logits,
X
xuezhong 已提交
5949
            'Labels': label,
X
xuezhong 已提交
5950 5951
            'CustomizedSamples': customized_samples,
            'CustomizedProbabilities': customized_probabilities
X
xuezhong 已提交
5952 5953 5954 5955
        },
        outputs={
            'Samples': samples,
            'Probabilities': probabilities,
X
xuezhong 已提交
5956
            'SampledLabels': sampled_label,
X
xuezhong 已提交
5957 5958 5959
            'SampledLogits': sampled_logits
        },
        attrs={
X
xuezhong 已提交
5960
            'use_customized_samples': use_customized_samples,
5961
            'uniq': True,
X
xuezhong 已提交
5962 5963 5964 5965
            'remove_accidental_hits': remove_accidental_hits,
            'num_samples': num_samples,
            'seed': seed
        })
X
xuezhong 已提交
5966 5967
    loss = helper.create_variable_for_type_inference(dtype=logits.dtype)
    softmax = helper.create_variable_for_type_inference(dtype=logits.dtype)
X
xuezhong 已提交
5968 5969 5970 5971 5972 5973
    helper.append_op(
        type='one_hot',
        inputs={'X': sampled_label},
        attrs={'depth': num_samples + 1},
        outputs={'Out': sampled_softlabel})

5974 5975
    helper.append_op(
        type='softmax_with_cross_entropy',
X
xuezhong 已提交
5976
        inputs={'Logits': sampled_logits,
X
xuezhong 已提交
5977
                'Label': sampled_softlabel},
X
xuezhong 已提交
5978 5979 5980
        outputs={'Softmax': softmax,
                 'Loss': loss},
        attrs={
X
xuezhong 已提交
5981
            'soft_label': True,
X
xuezhong 已提交
5982 5983 5984
            'ignore_index': False,
            'numeric_stable_mode': False
        })
X
xuezhong 已提交
5985
    return loss / num_true
X
xuezhong 已提交
5986 5987


5988 5989
def smooth_l1(x, y, inside_weight=None, outside_weight=None, sigma=None):
    """
Y
Yibing Liu 已提交
5990 5991
    This layer computes the smooth L1 loss for Variable :attr:`x` and :attr:`y`.
    It takes the first dimension of :attr:`x` and :attr:`y` as batch size.
Q
qingqing01 已提交
5992
    For each instance, it computes the smooth L1 loss element by element first
5993
    and then sums all the losses. So the shape of ouput Variable is
5994
    [batch_size, 1].
5995

5996 5997
    Args:
        x (Variable): A tensor with rank at least 2. The input value of smooth
Q
qingqing01 已提交
5998
            L1 loss op with shape [batch_size, dim1, ..., dimN].
5999
        y (Variable): A tensor with rank at least 2. The target value of smooth
Y
Yibing Liu 已提交
6000
            L1 loss op with same shape as :attr:`x`.
6001
        inside_weight (Variable|None):  A tensor with rank at least 2. This
6002 6003
            input is optional and should have same shape with :attr:`x`. If
            provided, the result of (:attr:`x` - :attr:`y`) will be multiplied
Y
Yibing Liu 已提交
6004
            by this tensor element by element.
6005
        outside_weight (Variable|None): A tensor with rank at least 2. This
6006 6007
            input is optional and should have same shape with :attr:`x`. If
            provided, the out smooth L1 loss will be multiplied by this tensor
Y
Yibing Liu 已提交
6008
            element by element.
6009
        sigma (float|None): Hyper parameter of smooth L1 loss layer. A float
6010 6011
           scalar with default value 1.0.

6012
    Returns:
6013
        Variable: The output smooth L1 loss with shape [batch_size, 1].
6014 6015 6016 6017 6018

    Examples:
        .. code-block:: python

            data = fluid.layers.data(name='data', shape=[128], dtype='float32')
F
stash  
fengjiayi 已提交
6019 6020
            label = fluid.layers.data(
                name='label', shape=[100], dtype='float32')
6021
            fc = fluid.layers.fc(input=data, size=100)
F
fengjiayi 已提交
6022
            out = fluid.layers.smooth_l1(x=fc, y=label)
6023
    """
6024

6025
    helper = LayerHelper('smooth_l1_loss', **locals())
X
Xin Pan 已提交
6026 6027
    diff = helper.create_variable_for_type_inference(dtype=x.dtype)
    loss = helper.create_variable_for_type_inference(dtype=x.dtype)
6028 6029 6030 6031 6032 6033 6034 6035 6036 6037 6038 6039
    helper.append_op(
        type='smooth_l1_loss',
        inputs={
            'X': x,
            'Y': y,
            'InsideWeight': inside_weight,
            'OutsideWeight': outside_weight
        },
        outputs={'Diff': diff,
                 'Out': loss},
        attrs={'sigma': sigma})
    return loss
6040 6041 6042 6043


def one_hot(input, depth):
    """
Y
Yibing Liu 已提交
6044
    This layer creates the one-hot representations for input indices.
6045 6046

    Args:
Y
Yibing Liu 已提交
6047 6048
        input(Variable): Input indices, last dimension must be 1.
        depth(scalar): An interger defining the depth of the one-hot dimension.
6049 6050

    Returns:
Y
Yibing Liu 已提交
6051
        Variable: The one-hot representations of input.
6052 6053

    Examples:
C
caoying03 已提交
6054
        .. code-block:: python
6055

Y
Yibing Liu 已提交
6056 6057
            label = layers.data(name="label", shape=[1], dtype="float32")
            one_hot_label = layers.one_hot(input=label, depth=10)
6058 6059
    """
    helper = LayerHelper("one_hot", **locals())
X
Xin Pan 已提交
6060
    one_hot_out = helper.create_variable_for_type_inference(dtype='float32')
6061 6062 6063 6064 6065 6066
    helper.append_op(
        type="one_hot",
        inputs={'X': input},
        attrs={'depth': depth},
        outputs={'Out': one_hot_out})
    return one_hot_out
Y
Yu Yang 已提交
6067 6068


Y
Yu Yang 已提交
6069
def autoincreased_step_counter(counter_name=None, begin=1, step=1):
Y
Yu Yang 已提交
6070
    """
Y
yi.wu 已提交
6071 6072 6073
    Create an auto-increase variable
    which will be automatically increased by 1 every mini-batch
    Return the run counter of the main program, default is started from 1.
Y
Yu Yang 已提交
6074 6075 6076 6077 6078 6079

    Args:
        counter_name(str): The counter name, default is '@STEP_COUNTER@'.
        begin(int): The first value of this counter.
        step(int): The increment step between each execution.

6080 6081
    Returns:
        Variable: The global run counter.
Y
yi.wu 已提交
6082 6083 6084 6085 6086 6087

    Examples:
        .. code-block:: python

           global_step = fluid.layers.autoincreased_step_counter(
               counter_name='@LR_DECAY_COUNTER@', begin=begin, step=1)
Y
Yu Yang 已提交
6088 6089
    """
    helper = LayerHelper('global_step_counter')
Y
Yu Yang 已提交
6090 6091
    if counter_name is None:
        counter_name = '@STEP_COUNTER@'
Y
Yu Yang 已提交
6092 6093 6094 6095 6096
    counter, is_new_var = helper.create_or_get_global_variable(
        name=counter_name, dtype='int64', shape=[1], persistable=True)
    if is_new_var:
        helper.set_variable_initializer(
            counter, initializer=Constant(
Y
Yu Yang 已提交
6097
                value=begin - 1, force_cpu=True))
W
Wu Yi 已提交
6098
        helper.main_program.global_block()._prepend_op(
Y
Yu Yang 已提交
6099 6100
            type='increment',
            inputs={'X': [counter]},
Y
Yu Yang 已提交
6101
            outputs={'Out': [counter]},
M
minqiyang 已提交
6102 6103
            attrs={'step': float(step)},
            stop_gradient=True)
Y
Yu Yang 已提交
6104 6105 6106
        counter.stop_gradient = True

    return counter
Y
yangyaming 已提交
6107 6108


6109
def reshape(x, shape, actual_shape=None, act=None, inplace=False, name=None):
C
caoying03 已提交
6110
    """
C
caoying03 已提交
6111 6112
    Gives a new shape to the input Tensor without changing its data.

6113 6114 6115 6116 6117
    The target shape can be given by :attr:`shape` or :attr:`actual_shape`.
    :attr:`shape` is a list of integer while :attr:`actual_shape` is a tensor
    variable. :attr:`actual_shape` has a higher priority than :attr:`shape`
    if it is provided, while :attr:`shape` still should be set correctly to
    gurantee shape inference in compile-time.
C
caoying03 已提交
6118

6119
    Some tricks exist when specifying the target shape.
C
caoying03 已提交
6120

6121 6122 6123 6124
    1. -1 means the value of this dimension is inferred from the total element
    number of x and remaining dimensions. Thus one and only one dimension can
    be set -1.

6125
    2. 0 means the actual dimension value is going to be copied from the
6126 6127 6128 6129
    corresponding dimension of x. The indice of 0s in shape can not exceed
    Rank(X).

    Here are some examples to explain it.
C
caoying03 已提交
6130 6131

    1. Given a 3-D tensor x with a shape [2, 4, 6], and the target shape
W
wanghaoshuang 已提交
6132
    is [6, 8], the reshape operator will transform x into a 2-D tensor with
6133
    shape [6, 8] and leaving x's data unchanged.
C
caoying03 已提交
6134

6135
    2. Given a 3-D tensor x with a shape [2, 4, 6], and the target shape
6136 6137
    specified is [2, 3, -1, 2], the reshape operator will transform x into a
    4-D tensor with shape [2, 3, 4, 2] and leaving x's data unchanged. In this
W
wanghaoshuang 已提交
6138 6139
    case, one dimension of the target shape is set to -1, the value of this
    dimension is inferred from the total element number of x and remaining
6140
    dimensions.
C
caoying03 已提交
6141

6142
    3. Given a 3-D tensor x with a shape [2, 4, 6], and the target shape
6143 6144 6145 6146
    is [-1, 0, 3, 2], the reshape operator will transform x into a 4-D tensor
    with shape [2, 4, 3, 2] and leaving x's data unchanged. In this case,
    besides -1, 0 means the actual dimension value is going to be copied from
    the corresponding dimension of x.
C
caoying03 已提交
6147 6148

    Args:
6149
        x(variable): The input tensor.
C
caoying03 已提交
6150 6151
        shape(list): The new shape. At most one dimension of the new shape can
                     be -1.
6152 6153 6154 6155 6156
        actual_shape(variable): An optional input. If provided, reshape
                                according to this given shape rather than
                                :attr:`shape` specifying shape. That is to
                                say :attr:`actual_shape` has a higher priority
                                than :attr:`shape`.
6157 6158
        act (str): The non-linear activation to be applied to the reshaped tensor
                   variable.
C
chengduozh 已提交
6159 6160 6161
        inplace(bool): If ``inplace`` is `True`, the input and output of ``layers.reshape``
                       are the same variable, otherwise, the input and output of
                       ``layers.reshape`` are different variables. Note that if :attr:`x`
C
chengduozh 已提交
6162
                       is more than one layer's input, ``inplace`` must be :attr:`False`.
6163
        name (str): The name of this layer. It is optional.
C
caoying03 已提交
6164

6165
    Returns:
G
guosheng 已提交
6166 6167 6168 6169
        Variable: The reshaped tensor variable if :attr:`act` is None. It is a \
                  new tensor variable if :attr:`inplace` is :attr:`False`, \
                  otherwise it is :attr:`x`. If :attr:`act` is not None, return \
                  the activated tensor variable.
C
caoying03 已提交
6170

X
Xin Pan 已提交
6171 6172 6173
    Raises:
        TypeError: if actual_shape is neither Variable nor None.

C
caoying03 已提交
6174 6175
    Examples:
        .. code-block:: python
G
guosheng 已提交
6176

6177
            data = fluid.layers.data(
6178
                name='data', shape=[2, 4, 6], dtype='float32')
C
caoying03 已提交
6179
            reshaped = fluid.layers.reshape(
G
guosheng 已提交
6180
                x=data, shape=[-1, 0, 3, 2], inplace=True)
C
caoying03 已提交
6181 6182 6183
    """

    if not (isinstance(shape, list) or isinstance(shape, tuple)):
L
luotao1 已提交
6184
        raise ValueError("Input shape must be a python list or tuple.")
X
Xin Pan 已提交
6185 6186 6187 6188 6189
    inputs = {"X": x}
    if isinstance(actual_shape, Variable):
        inputs["Shape"] = actual_shape
    elif actual_shape is not None:
        raise TypeError("actual_shape should either be Variable or None")
C
caoying03 已提交
6190

6191 6192 6193 6194 6195 6196 6197 6198 6199 6200 6201 6202 6203 6204 6205
    # Validate the shape
    unk_dim_idx = -1
    for dim_idx, dim_size in enumerate(shape):
        if dim_size == -1:
            assert unk_dim_idx == -1, (
                "Only one dimension in shape can be unknown.")
            unk_dim_idx = dim_idx
        elif dim_size == 0:
            assert dim_idx < len(x.shape), (
                "The indice of 0s in shape can not exceed Rank(X).")
        else:
            assert dim_size > 0, (
                "Each dimension size given in shape must not be negtive "
                "except one unknown dimension.")

6206
    helper = LayerHelper("reshape2", **locals())
6207 6208
    out = x if inplace else helper.create_variable_for_type_inference(
        dtype=x.dtype)
X
Xin Pan 已提交
6209
    x_shape = helper.create_variable_for_type_inference(dtype=x.dtype)
C
caoying03 已提交
6210
    helper.append_op(
6211
        type="reshape2",
X
Xin Pan 已提交
6212
        inputs=inputs,
D
dzhwinter 已提交
6213
        attrs={"shape": shape},
6214 6215
        outputs={"Out": out,
                 "XShape": x_shape})
C
caoying03 已提交
6216

D
dzhwinter 已提交
6217
    return helper.append_activation(out)
6218

6219

6220
def squeeze(input, axes, name=None):
Y
Yibing Liu 已提交
6221
    """
M
minqiyang 已提交
6222 6223 6224
    Remove single-dimensional entries from the shape of a tensor. Takes a
    parameter axes with a list of axes to squeeze. If axes is not provided, all
    the single dimensions will be removed from the shape. If an axis is
Y
Yibing Liu 已提交
6225
    selected with shape entry not equal to one, an error is raised.
M
minqiyang 已提交
6226

H
haowang101779990 已提交
6227 6228 6229 6230 6231 6232 6233 6234 6235 6236 6237 6238 6239 6240 6241 6242 6243 6244 6245 6246 6247
    For example:

    .. code-block:: text

        Case 1:

          Given
            X.shape = (1, 3, 1, 5)
          and
            axes = [0]
          we get:
            Out.shape = (3, 1, 5)

        Case 2:

          Given
            X.shape = (1, 3, 1, 5)
          and
            axes = []
          we get:
            Out.shape = (3, 5)
M
minqiyang 已提交
6248

Y
Yibing Liu 已提交
6249
    Args:
6250
        input (Variable): The input variable to be squeezed.
Y
Yibing Liu 已提交
6251
        axes (list): List of integers, indicating the dimensions to be squeezed.
6252
        name (str|None): Name for this layer.
Y
Yibing Liu 已提交
6253 6254 6255 6256 6257 6258 6259 6260

    Returns:
        Variable: Output squeezed variable.

    Examples:
        .. code-block:: python

            x = layers.data(name='x', shape=[5, 1, 10])
6261
            y = layers.sequeeze(input=x, axes=[1])
Y
Yibing Liu 已提交
6262 6263
    """
    helper = LayerHelper("squeeze", **locals())
X
Xin Pan 已提交
6264 6265
    out = helper.create_variable_for_type_inference(dtype=input.dtype)
    x_shape = helper.create_variable_for_type_inference(dtype=input.dtype)
Y
Yibing Liu 已提交
6266
    helper.append_op(
6267
        type="squeeze2",
6268
        inputs={"X": input},
Y
Yibing Liu 已提交
6269
        attrs={"axes": axes},
6270 6271
        outputs={"Out": out,
                 "XShape": x_shape})
Y
Yibing Liu 已提交
6272

6273 6274 6275
    return out


6276
def unsqueeze(input, axes, name=None):
Y
Yibing Liu 已提交
6277
    """
M
minqiyang 已提交
6278 6279 6280
    Insert single-dimensional entries to the shape of a tensor. Takes one
    required argument axes, a list of dimensions that will be inserted.
    Dimension indices in axes are as seen in the output tensor.
Y
Yibing Liu 已提交
6281

M
minqiyang 已提交
6282
    For example:
H
haowang101779990 已提交
6283 6284 6285

    .. code-block:: text

M
minqiyang 已提交
6286
      Given a tensor such that tensor with shape [3, 4, 5],
Y
Yibing Liu 已提交
6287
      then Unsqueezed tensor with axes=[0, 4] has shape [1, 3, 4, 5, 1].
M
minqiyang 已提交
6288

Y
Yibing Liu 已提交
6289
    Args:
6290
        input (Variable): The input variable to be unsqueezed.
Y
Yibing Liu 已提交
6291
        axes (list): List of integers, indicating the dimensions to be inserted.
6292
        name (str|None): Name for this layer.
Y
Yibing Liu 已提交
6293 6294 6295 6296 6297 6298 6299 6300

    Returns:
        Variable: Output unsqueezed variable.

    Examples:
        .. code-block:: python

            x = layers.data(name='x', shape=[5, 10])
6301
            y = layers.unsequeeze(input=x, axes=[1])
Y
Yibing Liu 已提交
6302 6303
    """
    helper = LayerHelper("unsqueeze", **locals())
X
Xin Pan 已提交
6304 6305
    out = helper.create_variable_for_type_inference(dtype=input.dtype)
    x_shape = helper.create_variable_for_type_inference(dtype=input.dtype)
Y
Yibing Liu 已提交
6306
    helper.append_op(
6307
        type="unsqueeze2",
6308
        inputs={"X": input},
Y
Yibing Liu 已提交
6309
        attrs={"axes": axes},
6310 6311
        outputs={"Out": out,
                 "XShape": x_shape})
Y
Yibing Liu 已提交
6312

6313 6314
    return out

6315

Y
yangyaming 已提交
6316
def lod_reset(x, y=None, target_lod=None):
Y
yangyaming 已提交
6317
    """
Y
Yibing Liu 已提交
6318
    Set LoD of :attr:`x` to a new one specified by :attr:`y` or
6319 6320 6321 6322
    :attr:`target_lod`. When :attr:`y` provided, :attr:`y.lod` would be
    considered as target LoD first, otherwise :attr:`y.data` would be
    considered as target LoD. If :attr:`y` is not provided, target LoD should
    be specified by :attr:`target_lod`. If target LoD is specified by
Y
Yibing Liu 已提交
6323
    :attr:`Y.data` or :attr:`target_lod`, only one level LoD is supported.
Y
yangyaming 已提交
6324 6325 6326 6327 6328 6329

    .. code-block:: text

        * Example 1:

            Given a 1-level LoDTensor x:
6330
                x.lod =  [[ 2,           3,                   1 ]]
Y
yangyaming 已提交
6331 6332 6333
                x.data = [[1.0], [2.0], [3.0], [4.0], [5.0], [6.0]]
                x.dims = [6, 1]

6334
            target_lod: [4, 2]
Y
yangyaming 已提交
6335 6336

            then we get a 1-level LoDTensor:
6337
                out.lod =  [[4,                          2]]
Y
yangyaming 已提交
6338 6339 6340 6341 6342 6343
                out.data = [[1.0], [2.0], [3.0], [4.0], [5.0], [6.0]]
                out.dims = [6, 1]

        * Example 2:

            Given a 1-level LoDTensor x:
6344
                x.lod =  [[2,            3,                   1]]
Y
yangyaming 已提交
6345 6346 6347 6348
                x.data = [[1.0], [2.0], [3.0], [4.0], [5.0], [6.0]]
                x.dims = [6, 1]

            y is a Tensor:
6349
                y.data = [[2, 4]]
Y
yangyaming 已提交
6350 6351 6352
                y.dims = [1, 3]

            then we get a 1-level LoDTensor:
6353
                out.lod =  [[2,            4]]
Y
yangyaming 已提交
6354 6355 6356 6357 6358 6359
                out.data = [[1.0], [2.0], [3.0], [4.0], [5.0], [6.0]]
                out.dims = [6, 1]

        * Example 3:

            Given a 1-level LoDTensor x:
6360
                x.lod =  [[2,            3,                   1]]
Y
yangyaming 已提交
6361 6362 6363 6364
                x.data = [[1.0], [2.0], [3.0], [4.0], [5.0], [6.0]]
                x.dims = [6, 1]

            y is a 2-level LoDTensor:
6365
                y.lod =  [[2, 2], [2, 2, 1, 1]]
Y
yangyaming 已提交
6366 6367 6368 6369
                y.data = [[1.1], [2.1], [3.1], [4.1], [5.1], [6.1]]
                y.dims = [6, 1]

            then we get a 2-level LoDTensor:
6370
                out.lod =  [[2, 2], [2, 2, 1, 1]]
Y
yangyaming 已提交
6371 6372 6373 6374 6375
                out.data = [[1.0], [2.0], [3.0], [4.0], [5.0], [6.0]]
                out.dims = [6, 1]

    Args:
        x (Variable): Input variable which could be a Tensor or LodTensor.
6376
        y (Variable|None): If provided, output's LoD would be derived
Y
Yibing Liu 已提交
6377
                           from :attr:`y`.
Y
yangyaming 已提交
6378
        target_lod (list|tuple|None): One level LoD which should be considered
Y
Yibing Liu 已提交
6379
                                      as target LoD when :attr:`y` not provided.
Y
yangyaming 已提交
6380 6381

    Returns:
Y
Yibing Liu 已提交
6382
        Variable: Output variable with LoD specified by this layer.
Y
yangyaming 已提交
6383 6384

    Raises:
Y
Yibing Liu 已提交
6385
        ValueError: If :attr:`y` and :attr:`target_lod` are both None.
Y
yangyaming 已提交
6386 6387 6388 6389 6390 6391 6392 6393 6394

    Examples:
        .. code-block:: python

            x = layers.data(name='x', shape=[10])
            y = layers.data(name='y', shape=[10, 20], lod_level=2)
            out = layers.lod_reset(x=x, y=y)
    """
    helper = LayerHelper("lod_reset", **locals())
X
Xin Pan 已提交
6395
    out = helper.create_variable_for_type_inference(dtype=x.dtype)
Y
yangyaming 已提交
6396 6397 6398 6399 6400 6401 6402 6403 6404 6405 6406 6407 6408 6409
    if y is not None:
        helper.append_op(
            type="lod_reset", inputs={'X': x,
                                      'Y': y}, outputs={'Out': out})
    elif target_lod is not None:
        helper.append_op(
            type="lod_reset",
            inputs={'X': x},
            attrs={'target_lod': target_lod},
            outputs={'Out': out})
    else:
        raise ValueError("y and target_lod should not be both None.")

    return out
D
dragonwarrior 已提交
6410 6411 6412 6413 6414 6415 6416 6417 6418 6419 6420


def lrn(input, n=5, k=1.0, alpha=1e-4, beta=0.75, name=None):
    """
    Local Response Normalization Layer. This layer performs a type of
    "lateral inhibition" by normalizing over local input regions.

    The formula is as follows:

    .. math::

D
dzhwinter 已提交
6421
      Output(i, x, y) = Input(i, x, y) / \\left(k + \\alpha \\sum\\limits^{\\min(C, c + n/2)}_{j = \\max(0, c - n/2)}(Input(j, x, y))^2\\right)^{\\beta}
D
dragonwarrior 已提交
6422 6423 6424 6425 6426 6427 6428 6429 6430 6431 6432 6433 6434 6435 6436 6437 6438 6439 6440 6441 6442 6443 6444 6445 6446 6447 6448 6449

    In the above equation:

    * :math:`n`: The number of channels to sum over.
    * :math:`k`: The offset (avoid being divided by 0).
    * :math:`alpha`: The scaling parameter.
    * :math:`beta`: The exponent parameter.

    Refer to `ImageNet Classification with Deep Convolutional Neural Networks
    <https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf>`_

    Args:
        input (Variable): The input tensor of this layer, and the dimension of input tensor must be 4.
        n (int, default 5): The number of channels to sum over.
        k (float, default 1.0): An offset (usually positive to avoid dividing by 0).
        alpha (float, default 1e-4): The scaling parameter.
        beta (float, default 0.75): The exponent.
        name (str, default None): A name for this operation.

    Raises:
        ValueError: If rank of the input tensor is not 4.

    Returns:
        A tensor variable storing the transformation result.

    Examples:
        .. code-block:: python

F
stash  
fengjiayi 已提交
6450 6451
          data = fluid.layers.data(
              name="data", shape=[3, 112, 112], dtype="float32")
D
dragonwarrior 已提交
6452 6453 6454 6455 6456 6457 6458 6459 6460 6461 6462 6463
          lrn = fluid.layers.lrn(input=data)
    """
    helper = LayerHelper('lrn', **locals())
    dtype = helper.input_dtype()
    input_shape = input.shape
    dims = len(input_shape)

    if dims != 4:
        raise ValueError(
            "dims of input must be 4(not %d), and it's order must be NCHW" %
            (dims))

X
Xin Pan 已提交
6464 6465 6466
    mid_out = helper.create_variable_for_type_inference(
        dtype=dtype, stop_gradient=True)
    lrn_out = helper.create_variable_for_type_inference(dtype)
D
dragonwarrior 已提交
6467 6468 6469 6470 6471 6472 6473 6474 6475 6476 6477 6478 6479
    helper.append_op(
        type="lrn",
        inputs={"X": input},
        outputs={
            "Out": lrn_out,
            "MidOut": mid_out,
        },
        attrs={"n": n,
               "k": k,
               "alpha": alpha,
               "beta": beta})

    return lrn_out
G
guosheng 已提交
6480 6481 6482 6483


def pad(x, paddings, pad_value=0., name=None):
    """
G
guosheng 已提交
6484
    Pads a tensor with a constant value given by :attr:`pad_value`, and the
W
wanghaoshuang 已提交
6485
    padded width is specified by :attr:`paddings`.
G
guosheng 已提交
6486

G
guosheng 已提交
6487 6488 6489 6490
    Specifically, the number of values padded before the contents of :attr:`x`
    in dimension :attr:`i` is indicated by :attr:`paddings[i]`, and the number
    of values padded after the contents of :attr:`x` in dimension :attr:`i` is
    indicated by :attr:`paddings[i+1]`.
G
guosheng 已提交
6491 6492 6493 6494 6495 6496 6497 6498 6499 6500 6501 6502 6503 6504 6505 6506 6507 6508 6509 6510 6511 6512

    See below for an example.

    .. code-block:: text

        Given:
            x = [[1, 2], [3, 4]]

            paddings = [0, 1, 1, 2]

            pad_value = 0

        Return:

            out = [[0, 1, 2, 0, 0]
                   [0, 3, 4, 0, 0]
                   [0, 0, 0, 0, 0]]

    Args:
        x (Variable): The input tensor variable.
        paddings (list): A list of integers. Its elements specify the padded
                         width before and after for each dimension in turn.
W
wanghaoshuang 已提交
6513
                         The length of :attr:paddings must be
G
guosheng 已提交
6514 6515 6516 6517 6518 6519 6520 6521 6522 6523
                         :math:`rank(x) \\times 2`.
        pad_value (float): The constant value used to pad.
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.

    Returns:
        Variable: The padded tensor variable.

    Examples:
        .. code-block:: python
G
guosheng 已提交
6524

G
guosheng 已提交
6525 6526 6527 6528 6529 6530
            # x is a rank 2 tensor variable.
            out = fluid.layers.pad(
                x=x, paddings=[0, 1, 1, 2], pad_value=0.)
    """
    helper = LayerHelper('pad', input=x, **locals())
    dtype = helper.input_dtype()
X
Xin Pan 已提交
6531
    out = helper.create_variable_for_type_inference(dtype)
G
guosheng 已提交
6532 6533 6534 6535 6536 6537 6538
    helper.append_op(
        type='pad',
        inputs={'X': x},
        outputs={'Out': out},
        attrs={'paddings': paddings,
               'pad_value': float(pad_value)})
    return out
6539 6540


C
chengduo 已提交
6541 6542 6543 6544 6545 6546 6547 6548 6549 6550 6551 6552 6553 6554 6555 6556 6557 6558 6559 6560 6561 6562 6563 6564 6565 6566 6567 6568 6569 6570 6571
def pad_constant_like(x, y, pad_value=0., name=None):
    """
    Pad input(Y) with :attr:`pad_value`, the number of values padded to
    the edges of each axis is specified by the difference of the shape
    of X and Y. ((0, shape_x_0 - shape_y_0), ... (0, shape_x_n - shape_y_n))
    unique pad widths for each axis. The input should be a k-D
    tensor(k > 0 and k < 7).

    See below for an example.

    .. code-block:: text

        Given:
            X = [[[[ 0,  1,  2],
                   [ 3,  4,  5]],
                  [[ 6,  7,  8],
                   [ 9, 10, 11]],
                  [[12, 13, 14],
                   [15, 16, 17]]],
                 [[[18, 19, 20],
                   [21, 22, 23]],
                  [[24, 25, 26],
                   [27, 28, 29]],
                  [[30, 31, 32],
                   [33, 34, 35]]]]
            X.shape = (2, 3, 2, 3)

            Y = [[[[35, 36, 37]],
                  [[38, 39, 40]],
                  [[41, 42, 43]]]]
            Y.shape = (1, 3, 1, 3)
T
Tink_Y 已提交
6572 6573
		And
            pad_value = -1,
C
chengduo 已提交
6574

T
Tink_Y 已提交
6575 6576 6577 6578 6579 6580 6581 6582 6583 6584 6585 6586 6587 6588
        Return:
            Out = [[[[35, 36, 37],
                     [-1, -1, -1]],
                    [[38, 39, 40],
                     [-1, -1, -1]],
                    [[41, 42, 43],
                     [-1, -1, -1]]],
                  [[[-1, -1, -1],
                    [-1, -1, -1]],
                   [[-1, -1, -1],
                    [-1, -1, -1]],
                   [[-1, -1, -1],
                    [-1, -1, -1]]]]
            Out.shape = (2, 3, 2, 3)
C
chengduo 已提交
6589 6590 6591 6592 6593 6594 6595 6596 6597 6598 6599 6600 6601 6602 6603 6604 6605 6606 6607 6608 6609

    Args:
        x (Variable): The input tensor variable.
        y (Variable): The input tensor variable.
        pad_value (float): The constant value used to pad.
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.

    Returns:
        Variable: The padded tensor variable.

    Examples:
        .. code-block:: python

            # x is a rank 4 tensor variable, x.shape = (2, 3, 2, 3)
            # y is a rank 4 tensor variable, y.shape = (1, 3, 1, 3)
            out = fluid.layers.pad_constant_like(x=x, y=y, pad_value=0.)
            # out is a rank 4 tensor variable, and out.shape = [2, 3 ,2 , 3]
    """
    helper = LayerHelper('pad_constant_like', input=x, **locals())
    dtype = helper.input_dtype()
X
Xin Pan 已提交
6610
    out = helper.create_variable_for_type_inference(dtype)
C
chengduo 已提交
6611 6612 6613 6614 6615 6616 6617 6618 6619
    helper.append_op(
        type='pad_constant_like',
        inputs={'X': x,
                'Y': y},
        outputs={'Out': out},
        attrs={'pad_value': float(pad_value)})
    return out


6620 6621 6622 6623 6624 6625 6626
def label_smooth(label,
                 prior_dist=None,
                 epsilon=0.1,
                 dtype="float32",
                 name=None):
    """
    Label smoothing is a mechanism to regularize the classifier layer and is
6627 6628
    called label-smoothing regularization (LSR).

6629 6630 6631 6632 6633 6634 6635 6636 6637 6638 6639 6640 6641 6642 6643 6644 6645 6646 6647 6648 6649 6650 6651
    Label smoothing is proposed to encourage the model to be less confident,
    since optimizing the log-likelihood of the correct label directly may
    cause overfitting and reduce the ability of the model to adapt. Label
    smoothing replaces the ground-truth label :math:`y` with the weighted sum
    of itself and some fixed distribution :math:`\mu`. For class :math:`k`,
    i.e.

    .. math::

        \\tilde{y_k} = (1 - \epsilon) * y_k + \epsilon * \mu_k,

    where :math:`1 - \epsilon` and :math:`\epsilon` are the weights
    respectively, and :math:`\\tilde{y}_k` is the smoothed label. Usually
    uniform distribution is used for :math:`\mu`.

    See more details about label smoothing in https://arxiv.org/abs/1512.00567.

    Args:
        label(Variable): The input variable containing the label data. The
                          label data should use one-hot representation.
        prior_dist(Variable): The prior distribution to be used to smooth
                              labels. If not provided, an uniform distribution
                              is used. The shape of :attr:`prior_dist` should
6652
                              be :math:`(1, class\_num)`.
6653 6654
        epsilon(float): The weight used to mix up the original ground-truth
                        distribution and the fixed distribution.
6655
        dtype(np.dtype|core.VarDesc.VarType|str): The type of data : float32,
6656 6657 6658 6659 6660 6661 6662 6663 6664 6665 6666 6667 6668 6669 6670 6671 6672 6673 6674
                                                  float_64, int etc.
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.

    Returns:
        Variable: The tensor variable containing the smoothed labels.

    Examples:
        .. code-block:: python

            label = layers.data(name="label", shape=[1], dtype="float32")
            one_hot_label = layers.one_hot(input=label, depth=10)
            smooth_label = layers.label_smooth(
                label=one_hot_label, epsilon=0.1, dtype="float32")
    """
    if epsilon > 1. or epsilon < 0.:
        raise ValueError("The value of epsilon must be between 0 and 1.")
    helper = LayerHelper("label_smooth", **locals())
    label.stop_gradient = True
X
Xin Pan 已提交
6675
    smooth_label = helper.create_variable_for_type_inference(dtype)
6676 6677 6678 6679 6680 6681 6682
    helper.append_op(
        type="label_smooth",
        inputs={"X": label,
                "PriorDist": prior_dist} if prior_dist else {"X": label},
        outputs={"Out": smooth_label},
        attrs={"epsilon": float(epsilon)})
    return smooth_label
6683 6684


W
wopeizl 已提交
6685 6686 6687 6688 6689 6690 6691 6692 6693 6694 6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711 6712 6713 6714 6715 6716 6717 6718 6719 6720
@templatedoc()
def roi_pool(input, rois, pooled_height=1, pooled_width=1, spatial_scale=1.0):
    """
    ${comment}

    Args:
        input (Variable): ${x_comment}
        rois (Variable): ROIs (Regions of Interest) to pool over.
        pooled_height (integer): ${pooled_height_comment} Default: 1
        pooled_width (integer): ${pooled_width_comment} Default: 1
        spatial_scale (float): ${spatial_scale_comment} Default: 1.0

    Returns:
        Variable: ${out_comment}.

    Examples:
        .. code-block:: python

            pool_out = fluid.layers.roi_pool(input=x, rois=rois, 7, 7, 1.0)
    """
    helper = LayerHelper('roi_pool', **locals())
    dtype = helper.input_dtype()
    pool_out = helper.create_variable_for_type_inference(dtype)
    argmaxes = helper.create_variable_for_type_inference(dtype='int32')
    helper.append_op(
        type="roi_pool",
        inputs={"X": input,
                "ROIs": rois},
        outputs={"Out": pool_out,
                 "Argmax": argmaxes},
        attrs={
            "pooled_height": pooled_height,
            "pooled_width": pooled_width,
            "spatial_scale": spatial_scale
        })
    return pool_out
W
whs 已提交
6721 6722


J
jerrywgz 已提交
6723 6724 6725 6726 6727 6728
@templatedoc()
def roi_align(input,
              rois,
              pooled_height=1,
              pooled_width=1,
              spatial_scale=1.0,
J
jerrywgz 已提交
6729 6730
              sampling_ratio=-1,
              name=None):
J
jerrywgz 已提交
6731 6732 6733 6734 6735 6736 6737 6738 6739 6740 6741 6742 6743 6744 6745 6746
    """
    ${comment}

    Args:
        input (Variable): ${x_comment}
        rois (Variable): ROIs (Regions of Interest) to pool over.
        pooled_height (integer): ${pooled_height_comment} Default: 1
        pooled_width (integer): ${pooled_width_comment} Default: 1
        spatial_scale (float): ${spatial_scale_comment} Default: 1.0
        sampling_ratio(intger): ${sampling_ratio_comment} Default: -1

    Returns:
        Variable: ${out_comment}.
    Examples:
        .. code-block:: python

6747 6748 6749
            align_out = fluid.layers.roi_align(input=x,
                                               rois=rois,
                                               pooled_height=7,
J
jerrywgz 已提交
6750 6751 6752 6753 6754 6755
                                               pooled_width=7,
                                               spatial_scale=0.5,
                                               sampling_ratio=-1)
    """
    helper = LayerHelper('roi_align', **locals())
    dtype = helper.input_dtype()
X
Xin Pan 已提交
6756
    align_out = helper.create_variable_for_type_inference(dtype)
J
jerrywgz 已提交
6757 6758 6759 6760 6761 6762 6763 6764 6765 6766 6767 6768 6769 6770
    helper.append_op(
        type="roi_align",
        inputs={"X": input,
                "ROIs": rois},
        outputs={"Out": align_out},
        attrs={
            "pooled_height": pooled_height,
            "pooled_width": pooled_width,
            "spatial_scale": spatial_scale,
            "sampling_ratio": sampling_ratio
        })
    return align_out


W
whs 已提交
6771 6772 6773 6774 6775 6776 6777 6778 6779 6780 6781 6782 6783 6784 6785 6786 6787 6788 6789 6790 6791 6792 6793 6794 6795 6796
def dice_loss(input, label, epsilon=0.00001):
    """
    Dice loss for comparing the similarity of two batch of data,
    usually is used for binary image segmentation i.e. labels are binary.
    The dice loss can be defined as below equation:

    .. math::

        dice\_loss &= 1 - \\frac{2 * intersection\_area}{total\_area} \\\\
                  &= \\frac{(total\_area - intersection\_area) - intersection\_area}{total\_area} \\\\
                  &= \\frac{(union\_area - intersection\_area)}{total\_area}


    Args:
        input (Variable): The predictions with rank>=2. The first dimension is batch size,
                          and the last dimension is class number.
        label (Variable): The groud truth with the same rank with input. The first dimension
                          is batch size, and the last dimension is 1.
        epsilon (float): The epsilon will be added to the numerator and denominator.
                         If both input and label are empty, it makes sure dice is 1.
                         Default: 0.00001

    Returns:
        dice_loss (Variable): The dice loss with shape [1].

    Examples:
6797 6798
        .. code-block:: python

W
whs 已提交
6799 6800 6801 6802
            predictions = fluid.layers.softmax(x)
            loss = fluid.layers.dice_loss(input=predictions, label=label, 2)
    """
    label = one_hot(label, depth=input.shape[-1])
6803
    reduce_dim = list(range(1, len(input.shape)))
W
whs 已提交
6804 6805 6806 6807 6808 6809
    inse = reduce_sum(input * label, dim=reduce_dim)
    dice_denominator = reduce_sum(
        input, dim=reduce_dim) + reduce_sum(
            label, dim=reduce_dim)
    dice_score = 1 - inse * 2 / (dice_denominator + epsilon)
    return reduce_mean(dice_score)
6810 6811


6812 6813 6814 6815
def image_resize(input,
                 out_shape=None,
                 scale=None,
                 name=None,
6816
                 resample='BILINEAR',
6817 6818
                 actual_shape=None,
                 align_corners=True,
T
tink2123 已提交
6819
                 align_mode=1):
6820
    """
Q
qiaolongfei 已提交
6821
    **Resize a Batch of Images**
F
stash  
fengjiayi 已提交
6822

6823
    The input must be a tensor of the shape (num_batches, channels, in_h, in_w),
6824 6825 6826
    and the resizing only applies on the last two dimensions(hight and width).

    Supporting resample methods:
Q
update  
qiaolongfei 已提交
6827

6828
        'BILINEAR' : Bilinear interpolation
T
Tink_Y 已提交
6829

6830
        'NEAREST' : Nearest neighbor interpolation
F
stash  
fengjiayi 已提交
6831

6832 6833 6834 6835 6836 6837 6838 6839 6840 6841
    Nearest neighbor interpolation is to perform nearest neighbor interpolation
    in both the 3rd dimention(in height direction) and the 4th dimention(in width 
    direction) on input tensor.
            
    Bilinear interpolation is an extension of linear interpolation for 
    interpolating functions of two variables (e.g. H-direction and 
    W-direction in this op) on a rectilinear 2D grid. The key idea is 
    to perform linear interpolation first in one direction, and then 
    again in the other direction.

T
tink2123 已提交
6842
    Align_corners and align_mode are optinal parameters,the calculation method 
6843 6844 6845 6846
    of interpolation can be selected by them.

    Example:

T
tink2123 已提交
6847
      For scale:
6848 6849 6850 6851 6852 6853 6854 6855 6856 6857 6858 6859
      
        if align_corners = True && out_size > 1 :

          scale_factor = (in_size-1.0)/(out_size-1.0)
        
        else:
          
          scale_factor = float(in_size/out_size)
        
      
      Nearest neighbor interpolation:
      
T
tink2123 已提交
6860
      if:
6861 6862 6863 6864 6865 6866 6867 6868
          align_corners = False

          input : (N,C,H_in,W_in)
          output: (N,C,H_out,W_out) where:

          H_out = \left \lfloor {H_{in} * scale_{}factor}} \right \rfloor
          W_out = \left \lfloor {W_{in} * scale_{}factor}} \right \rfloor

T
tink2123 已提交
6869
      else:
6870 6871 6872 6873 6874 6875 6876 6877 6878 6879
          align_corners = True

          input : (N,C,H_in,W_in)
          output: (N,C,H_out,W_out) where:

          H_out = round(H_{in} * scale_{factor})
          W_out = round(W_{in} * scale_{factor})

      Bilinear interpolation:

T
tink2123 已提交
6880
      if:
6881 6882 6883 6884 6885 6886 6887 6888 6889
          align_corners = False , align_mode = 0
          
          input : (N,C,H_in,W_in)
          output: (N,C,H_out,W_out) where:
          
          H_out = (H_{in}+0.5) * scale_{factor} - 0.5
          W_out = (W_{in}+0.5) * scale_{factor} - 0.5


T
tink2123 已提交
6890
      else:
6891 6892 6893 6894 6895 6896 6897 6898 6899 6900 6901 6902 6903 6904 6905
       
          input : (N,C,H_in,W_in)
          output: (N,C,H_out,W_out) where:

          H_out = H_{in} * scale_{factor}
          W_out = W_{in} * scale_{factor}

    For details of nearest neighbor interpolation, please refer to Wikipedia: 
    https://en.wikipedia.org/wiki/Nearest-neighbor_interpolation.

    For details of bilinear interpolation, please refer to Wikipedia: 
    https://en.wikipedia.org/wiki/Bilinear_interpolation.



6906
    Args:
6907
        input (Variable): The input tensor of image resize layer,
6908 6909
                          This is a 4-D tensor of the shape
                          (num_batches, channels, in_h, in_w).
6910
        out_shape(list|tuple|Variable|None): Output shape of image resize
6911 6912
                                    layer, the shape is (out_h, out_w).
                                    Default: None
B
baiyf 已提交
6913
        scale(float|None): The multiplier for the input height or width.
6914 6915 6916
                         At least one of out_shape or scale must be set.
                         And out_shape has a higher priority than scale.
                         Default: None
6917 6918
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.
6919
        resample(str): The resample method. It supports 'BILINEAR' and 'NEAREST'
6920
                       currently.
6921
                       Default: 'BILINEAR'
6922 6923 6924
        actual_shape(Variable): An optional input to specify output shape
                                dynamically. If provided, image resize
                                according to this given shape rather than
6925
                                :attr:`out_shape` and :attr:`scale` specifying
6926 6927 6928 6929 6930 6931 6932
                                shape. That is to say actual_shape has the
                                highest priority. It is recommended to use
                                actual_shape instead of :attr:`out_shape` if you
                                want to specify output shape dynamically. When
                                using actual_shape to specify output shape, one of
                                :attr:`out_shape` and :attr:`scale` should also be
                                set, otherwise errors would be occured in graph
6933 6934
                                constructing stage.
                                Default: None
6935 6936 6937 6938
        align_corners(bool) :  An optional bool, If True, the centers of the 4 corner pixels of the 
                               input and output tensors are aligned, preserving the values at the 
                               corner pixels.
                               Default: True
T
tink2123 已提交
6939
        align_mode(int)  :  An optional for bilinear interpolation. can be \'0\' 
T
tink2123 已提交
6940 6941
                            for src_idx = scale*(dst_indx+0.5)-0.5 , can be \'1\' for 
                            src_idx = scale*dst_index .
6942 6943

    Returns:
Q
update  
qiaolongfei 已提交
6944 6945
        Variable: The output is a 4-D tensor of the shape
        (num_batches, channls, out_h, out_w).
F
stash  
fengjiayi 已提交
6946

6947 6948 6949
    Raises:
        TypeError: out_shape should be a list or tuple or Variable.
        TypeError: actual_shape should either be Variable or None.
6950
        ValueError: The 'resample' of image_resize can only be 'BILINEAR'
6951 6952 6953
                    or 'NEAREST' currently.
        ValueError: One of out_shape and scale must not be None.
        ValueError: out_shape length should be 2.
6954 6955
        TypeError: align_corners shoule be a bool value
        ValueError: align_mode can only be '0' or '1'
6956

6957 6958 6959
    Examples:
        .. code-block:: python

6960
            out = fluid.layers.image_resize(input, out_shape=[12, 12], resample="NEAREST")
6961
    """
6962 6963 6964 6965
    resample_methods = {
        'BILINEAR': 'bilinear',
        'NEAREST': 'nearest',
    }
6966 6967
    if resample not in resample_methods:
        raise ValueError(
6968
            "The 'resample' of image_resize can only be 'BILINEAR' or 'NEAREST' currently."
6969
        )
6970
    resample_type = resample_methods[resample]
6971 6972 6973 6974 6975 6976

    if not isinstance(align_corners, bool):
        raise TypeError("Attr align_corners should be a bool value")
    if align_mode != 0 and align_mode != 1:
        raise ValueError("align_mode can only be 0 or 1")

6977
    if out_shape is None and scale is None:
6978
        raise ValueError("One of out_shape and scale must not be None.")
6979
    helper = LayerHelper('{}_interp'.format(resample_type), **locals())
6980
    dtype = helper.input_dtype()
6981 6982 6983 6984

    def _is_list_or_turple_(data):
        return (isinstance(data, list) or isinstance(data, tuple))

6985 6986 6987
    out_h = 0
    out_w = 0
    inputs = {"X": input}
6988
    if out_shape is not None:
6989 6990 6991 6992
        if isinstance(out_shape, Variable):
            warnings.warn("out_shape as Variable type is deprecated, \
                    it is recommended to use actual_shape instead of \
                    out_shape to specify output shape dynamically.")
6993
            inputs['OutSize'] = out_shape
6994 6995 6996 6997 6998 6999 7000 7001
        elif not (_is_list_or_turple_(out_shape)):
            raise TypeError("out_shape should be a list or tuple or Variable.")
        elif len(out_shape) != 2:
            raise ValueError("out_shape length should be 2.")

        out_shape = list(map(int, out_shape))
        out_h = out_shape[0]
        out_w = out_shape[1]
7002 7003 7004 7005
    else:
        out_h = int(input.shape[2] * scale)
        out_w = int(input.shape[3] * scale)

7006 7007 7008 7009 7010
    if isinstance(actual_shape, Variable):
        inputs["OutSize"] = actual_shape
    elif actual_shape is not None:
        raise TypeError("actual_shape should either be Variable or None.")

X
Xin Pan 已提交
7011
    out = helper.create_variable_for_type_inference(dtype)
7012
    helper.append_op(
7013
        type='{}_interp'.format(resample_type),
7014
        inputs=inputs,
7015
        outputs={"Out": out},
7016 7017 7018 7019 7020 7021 7022
        attrs={
            "out_h": out_h,
            "out_w": out_w,
            "interp_method": resample_type,
            "align_corners": align_corners,
            "align_mode": align_mode
        })
7023
    return out
F
stash  
fengjiayi 已提交
7024 7025


7026
@templatedoc(op_type="bilinear_interp")
7027 7028 7029 7030
def resize_bilinear(input,
                    out_shape=None,
                    scale=None,
                    name=None,
7031 7032
                    actual_shape=None,
                    align_corners=True,
T
tink2123 已提交
7033
                    align_mode=1):
7034
    """
7035 7036
    Resize input by performing bilinear interpolation based on given
    output shape which specified by actual_shape, out_shape and scale
7037 7038
    in priority order.

7039 7040 7041 7042
    Bilinear interpolation is an extension of linear interpolation for
    interpolating functions of two variables (e.g. H-direction and
    W-direction in this op) on a rectilinear 2D grid. The key idea is
    to perform linear interpolation first in one direction, and then
7043 7044
    again in the other direction.

7045
    For details of bilinear interpolation, please refer to Wikipedia:
7046
    https://en.wikipedia.org/wiki/Bilinear_interpolation
Y
yuyang18 已提交
7047

T
tink2123 已提交
7048
    Align_corners and align_mode are optinal parameters,the calculation 
7049 7050 7051
    method of interpolation can be selected by them.


T
tink2123 已提交
7052
    Align_corners and align_mode are optinal parameters,the calculation method 
7053 7054 7055 7056
    of interpolation can be selected by them.

    Example:

T
tink2123 已提交
7057
      For scale:
7058 7059 7060 7061 7062 7063 7064 7065 7066 7067 7068
      
        if align_corners = True && out_size > 1 :

          scale_factor = (in_size-1.0)/(out_size-1.0)
        
        else:
          
          scale_factor = float(in_size/out_size)     

    Bilinear interpolation:

T
tink2123 已提交
7069
      if:
7070 7071 7072 7073 7074 7075 7076 7077 7078
          align_corners = False , align_mode = 0
          
          input : (N,C,H_in,W_in)
          output: (N,C,H_out,W_out) where:
          
          H_out = (H_{in}+0.5) * scale_{factor} - 0.5
          W_out = (W_{in}+0.5) * scale_{factor} - 0.5


T
tink2123 已提交
7079 7080
      else:

7081 7082 7083 7084 7085 7086 7087 7088
          input : (N,C,H_in,W_in)
          output: (N,C,H_out,W_out) where:

          H_out = H_{in} * scale_{factor}
          W_out = W_{in} * scale_{factor}



Y
yuyang18 已提交
7089 7090 7091 7092
    Args:
        input(${x_type}): ${x_comment}.

        out_shape(${out_size_type}): ${out_size_comment}.
7093

Y
yuyang18 已提交
7094 7095 7096 7097 7098
        scale(float|None): The multiplier for the input height or width. At
             least one of out_shape or scale must be set. And out_shape has
             a higher priority than scale. Default: None.

        name(str|None): The output variable name.
7099 7100 7101
        actual_shape(Variable): An optional input to specify output shape
                                dynamically. If provided, image resize
                                according to this given shape rather than
7102
                                :attr:`out_shape` and :attr:`scale` specifying
7103 7104 7105 7106 7107 7108 7109
                                shape. That is to say actual_shape has the
                                highest priority. It is recommended to use
                                actual_shape instead of :attr:`out_shape` if you
                                want to specify output shape dynamically. When
                                using actual_shape to specify output shape, one of
                                :attr:`out_shape` and :attr:`scale` should also be
                                set, otherwise errors would be occured in graph
7110 7111
                                constructing stage.
                                Default: None
7112 7113
        align_corners(bool): ${align_corners_comment}
        align_mode(bool): ${align_mode_comment}
Y
yuyang18 已提交
7114 7115 7116

    Returns:
        ${out_comment}.
7117 7118 7119 7120 7121

    Examples:
        .. code-block:: python

            out = fluid.layers.resize_bilinear(input, out_shape=[12, 12])
7122 7123
    """

7124 7125
    return image_resize(input, out_shape, scale, name, 'BILINEAR', actual_shape,
                        align_corners, align_mode)
7126 7127


7128
@templatedoc(op_type="nearest_interp")
7129 7130 7131 7132
def resize_nearest(input,
                   out_shape=None,
                   scale=None,
                   name=None,
7133 7134
                   actual_shape=None,
                   align_corners=True):
7135
    """
7136
    Resize input by performing nearest neighbor interpolation in both the
7137 7138
    3rd dimention(in height direction) and the 4th dimention(in width
    direction) based on given output shape which specified by actual_shape,
7139 7140
    out_shape and scale in priority order.

7141 7142
    Example:

T
tink2123 已提交
7143
      For scale:
7144 7145 7146 7147 7148 7149 7150 7151 7152 7153 7154 7155
      
        if align_corners = True && out_size > 1 :

          scale_factor = (in_size-1.0)/(out_size-1.0)
        
        else:
          
          scale_factor = float(in_size/out_size)
        
      
      Nearest neighbor interpolation:
      
T
tink2123 已提交
7156
      if:
7157 7158 7159 7160 7161 7162 7163 7164
          align_corners = False

          input : (N,C,H_in,W_in)
          output: (N,C,H_out,W_out) where:

          H_out = \left \lfloor {H_{in} * scale_{}factor}} \right \rfloor
          W_out = \left \lfloor {W_{in} * scale_{}factor}} \right \rfloor

T
tink2123 已提交
7165
      else:
7166 7167 7168 7169 7170 7171 7172 7173 7174
          align_corners = True

          input : (N,C,H_in,W_in)
          output: (N,C,H_out,W_out) where:

          H_out = round(H_{in} * scale_{factor})
          W_out = round(W_{in} * scale_{factor})


7175
    For details of nearest neighbor interpolation, please refer to Wikipedia:
7176
    https://en.wikipedia.org/wiki/Nearest-neighbor_interpolation
Y
yuyang18 已提交
7177 7178 7179 7180 7181

    Args:
        input(${x_type}): ${x_comment}.

        out_shape(${out_size_type}): ${out_size_comment}.
7182

Y
yuyang18 已提交
7183 7184 7185 7186 7187
        scale(float|None): The multiplier for the input height or width. At
             least one of out_shape or scale must be set. And out_shape has
             a higher priority than scale. Default: None.

        name(str|None): The output variable name.
7188 7189 7190
        actual_shape(Variable): An optional input to specify output shape
                                dynamically. If provided, image resize
                                according to this given shape rather than
7191
                                :attr:`out_shape` and :attr:`scale` specifying
7192 7193 7194 7195 7196 7197 7198
                                shape. That is to say actual_shape has the
                                highest priority. It is recommended to use
                                actual_shape instead of :attr:`out_shape` if you
                                want to specify output shape dynamically. When
                                using actual_shape to specify output shape, one of
                                :attr:`out_shape` and :attr:`scale` should also be
                                set, otherwise errors would be occured in graph
7199 7200
                                constructing stage.
                                Default: None
7201
        align_corners(bool): ${align_corners_comment}
Y
yuyang18 已提交
7202 7203 7204

    Returns:
        ${out_comment}.
7205 7206 7207 7208 7209

    Examples:
        .. code-block:: python

            out = fluid.layers.resize_nearest(input, out_shape=[12, 12])
7210 7211
    """

7212 7213
    return image_resize(input, out_shape, scale, name, 'NEAREST', actual_shape,
                        align_corners)
7214 7215 7216 7217


def image_resize_short(input, out_short_len, resample='BILINEAR'):
    """
7218 7219 7220
    Resize a batch of images. The short edge of input images will be
    resized to the given 'out_short_len'. The long edge of input images
    will be resized proportionately to make images' length-width ratio
7221 7222 7223 7224 7225 7226 7227
    constant.

    Args:
        input (Variable): The input tensor of image resize layer,
                          This is a 4-D tensor of the shape
                          (num_batches, channels, in_h, in_w).
        out_short_len(int): The length of output images' short edge.
7228
        resample (str): resample method, default: BILINEAR.
F
fengjiayi 已提交
7229

7230
    Returns:
Q
update  
qiaolongfei 已提交
7231
        Variable: The output is a 4-D tensor of the shape
7232
        (num_batches, channls, out_h, out_w).
7233 7234 7235 7236 7237 7238 7239 7240 7241 7242
    """
    in_shape = input.shape
    if len(in_shape) != 4:
        raise ValueError(
            "The rank of input must be 4 (num_batches, channels, in_h, in_w).")
    hw = in_shape[2:4]
    short_idx = hw.index(min(hw))
    long_idx = 1 - short_idx
    out_shape = list(hw)
    out_shape[short_idx] = out_short_len
F
fengjiayi 已提交
7243 7244 7245
    out_shape[long_idx] = int(
        float(out_shape[long_idx]) * (float(out_short_len) / float(hw[
            short_idx])) + 0.5)
7246 7247 7248
    return image_resize(input=input, out_shape=out_shape, resample=resample)


W
whs 已提交
7249 7250
def gather(input, index):
    """
Q
qiaolongfei 已提交
7251 7252
    **Gather Layer**

7253
    Output is obtained by gathering entries of the outer-most dimension
W
whs 已提交
7254 7255 7256 7257
    of X indexed by `index` and concatenate them together.

    .. math::

7258
        Out = X[Index]
W
whs 已提交
7259 7260 7261 7262 7263 7264 7265


    .. code-block:: text


                Given:

7266 7267
                X = [[1, 2],
                     [3, 4],
W
whs 已提交
7268 7269 7270 7271 7272 7273 7274 7275 7276 7277
                     [5, 6]]

                Index = [1, 2]

                Then:

                Out = [[3, 4],
                       [5, 6]]

    Args:
7278
        input (Variable): The source input with rank>=1.
W
whs 已提交
7279 7280 7281 7282 7283 7284
        index (Variable): The index input with rank=1.

    Returns:
        output (Variable): The output is a tensor with the same rank as input.

    Examples:
W
whs 已提交
7285

W
whs 已提交
7286 7287 7288 7289 7290 7291
        .. code-block:: python

            output = fluid.layers.gather(x, index)
    """
    helper = LayerHelper('gather', **locals())
    dtype = helper.input_dtype()
X
Xin Pan 已提交
7292
    out = helper.create_variable_for_type_inference(dtype)
W
whs 已提交
7293 7294 7295 7296 7297 7298 7299 7300
    helper.append_op(
        type="gather",
        inputs={"X": input,
                "Index": index},
        outputs={"Out": out})
    return out


7301 7302 7303 7304 7305 7306 7307 7308 7309 7310 7311 7312 7313 7314 7315 7316 7317 7318 7319 7320 7321 7322 7323 7324 7325 7326 7327 7328 7329 7330 7331
def scatter(input, index, updates, name=None):
    """
    **Scatter Layer**

    Output is obtained by updating the input on selected indices on the first
    axis.

    .. math::

        Out = X
        Out[Ids] = Updates

    Args:
        input (Variable): The source input with rank>=1.
        index (Variable): The index input with rank=1. Its dtype should be
                          int32 or int64 as it is used as indexes.
        updates (Variable): The updated value of scatter op.
        name (str|None): The output variable name. Default None.

    Returns:
        output (Variable): The output is a tensor with the same shape as input.

    Examples:

        .. code-block:: python

            output = fluid.layers.scatter(input, index, updates)

    """
    helper = LayerHelper('scatter', **locals())
    dtype = helper.input_dtype()
X
Xin Pan 已提交
7332
    out = helper.create_variable_for_type_inference(dtype)
7333 7334 7335 7336 7337 7338 7339 7340 7341
    helper.append_op(
        type="scatter",
        inputs={"X": input,
                "Ids": index,
                "Updates": updates},
        outputs={"Out": out})
    return out


Q
Qingsheng Li 已提交
7342 7343 7344 7345 7346 7347 7348 7349 7350
def sequence_scatter(input, index, updates, name=None):
    """
    **Sequence Scatter Layer**

    This operator scatters the Updates tensor to the input X. It uses the LoD
    information of Ids to select the rows to update, and use the values in Ids as
    the columns to update in each row of X.

    Here is an example:
H
haowang101779990 已提交
7351

Q
Qingsheng Li 已提交
7352
    Given the following input:
H
haowang101779990 已提交
7353

Q
Qingsheng Li 已提交
7354
    .. code-block:: text
H
haowang101779990 已提交
7355

Q
Qingsheng Li 已提交
7356 7357 7358 7359 7360 7361 7362 7363 7364 7365 7366 7367
        input.data = [[1.0, 1.0, 1.0, 1.0, 1.0, 1.0],
                      [1.0, 1.0, 1.0, 1.0, 1.0, 1.0],
                      [1.0, 1.0, 1.0, 1.0, 1.0, 1.0]]
        input.dims = [3, 6]

        index.data = [[0], [1], [2], [5], [4], [3], [2], [1], [3], [2], [5], [4]]
        index.lod =  [[0,        3,                       8,                 12]]

        updates.data = [[0.3], [0.3], [0.4], [0.1], [0.2], [0.3], [0.4], [0.0], [0.2], [0.3], [0.1], [0.4]]
        updates.lod =  [[  0,            3,                                 8,                         12]]

    Then we have the output:
H
haowang101779990 已提交
7368

Q
Qingsheng Li 已提交
7369
    .. code-block:: text
H
haowang101779990 已提交
7370

Q
Qingsheng Li 已提交
7371 7372 7373 7374 7375 7376 7377 7378 7379 7380 7381 7382 7383 7384 7385
        out.data = [[1.3, 1.3, 1.4, 1.0, 1.0, 1.0],
                    [1.0, 1.0, 1.4, 1.3, 1.2, 1.1],
                    [1.0, 1.0, 1.3, 1.2, 1.4, 1.1]]
        out.dims = X.dims = [3, 6]

    Args:
        input (Variable): The source input with rank>=1.
        index (Variable): A LoD Tensor. The index input of sequence scatter op
            where input will be  updated. The index input with rank=1. Its dtype
            should be int32 or int64 as it is used as indexes.
        updates (Variable): A LoD Tensor. The values to scatter to the input
            tensor X, must be a LoDTensor with the same LoD information as index.
        name (str|None): The output variable name. Default None.

    Returns:
H
haowang101779990 已提交
7386
        Variable: The output is a tensor with the same shape as input.
Q
Qingsheng Li 已提交
7387 7388 7389 7390 7391 7392 7393 7394 7395 7396

    Examples:

        .. code-block:: python

            output = fluid.layers.sequence_scatter(input, index, updates)

    """
    helper = LayerHelper('sequence_scatter', **locals())
    dtype = helper.input_dtype()
X
Xin Pan 已提交
7397
    out = helper.create_variable_for_type_inference(dtype)
Q
Qingsheng Li 已提交
7398 7399 7400 7401 7402 7403 7404 7405 7406
    helper.append_op(
        type="sequence_scatter",
        inputs={"X": input,
                "Ids": index,
                "Updates": updates},
        outputs={"Out": out})
    return out


Y
yuyang18 已提交
7407 7408 7409 7410 7411 7412 7413 7414 7415 7416 7417 7418 7419
@templatedoc()
def random_crop(x, shape, seed=None):
    """
    ${comment}

    Args:
        x(${x_type}): ${x_comment}
        shape(${shape_type}): ${shape_comment}
        seed(int|${seed_type}|None): ${seed_comment} By default, the seed will
            get from `random.randint(-65536, 65535)`.

    Returns:
        ${out_comment}
7420

7421 7422 7423
    Examples:
        >>> img = fluid.layers.data("img", [3, 256, 256])
        >>> cropped_img = fluid.layers.random_crop(img, shape=[3, 224, 224])
Y
yuyang18 已提交
7424
    """
F
stash  
fengjiayi 已提交
7425
    helper = LayerHelper("random_crop", **locals())
F
fengjiayi 已提交
7426
    dtype = x.dtype
X
Xin Pan 已提交
7427
    out = helper.create_variable_for_type_inference(dtype)
Y
yuyang18 已提交
7428
    if seed is None:
7429
        seed = np.random.randint(-65536, 65536)
F
fengjiayi 已提交
7430
    op_attrs = {"shape": shape}
F
stash  
fengjiayi 已提交
7431
    if isinstance(seed, int):
F
fengjiayi 已提交
7432 7433 7434 7435 7436
        op_attrs["startup_seed"] = seed
        seed = helper.create_variable(
            name=unique_name.generate("random_crop_seed"),
            dtype="int64",
            persistable=True)
F
stash  
fengjiayi 已提交
7437 7438 7439 7440
    elif not isinstance(seed, Variable):
        raise ValueError("'seed' must be a Variable or an int.")
    helper.append_op(
        type="random_crop",
F
fix  
fengjiayi 已提交
7441
        inputs={"X": x,
F
stash  
fengjiayi 已提交
7442 7443
                "Seed": seed},
        outputs={"Out": out,
F
fengjiayi 已提交
7444 7445
                 "SeedOut": seed},
        attrs=op_attrs)
F
stash  
fengjiayi 已提交
7446
    return out
W
whs 已提交
7447 7448


7449
def log(x, name=None):
W
wanghaoshuang 已提交
7450 7451 7452 7453 7454
    """
    Calculates the natural log of the given input tensor, element-wise.

    .. math::

7455
        Out = \\ln(x)
W
wanghaoshuang 已提交
7456 7457

    Args:
7458
        x (Variable): Input tensor.
7459 7460
        name (str|None, default None): A name for this layer If set None,
            the layer will be named automatically.
W
wanghaoshuang 已提交
7461 7462 7463 7464 7465 7466 7467 7468

    Returns:
        Variable: The natural log of the input tensor computed element-wise.

    Examples:

        .. code-block:: python

7469
            output = fluid.layers.log(x)
W
wanghaoshuang 已提交
7470 7471
    """
    helper = LayerHelper('log', **locals())
W
wanghaoshuang 已提交
7472
    dtype = helper.input_dtype(input_param_name='x')
X
Xin Pan 已提交
7473
    out = helper.create_variable_for_type_inference(dtype)
W
wanghaoshuang 已提交
7474
    helper.append_op(type="log", inputs={"X": x}, outputs={"Out": out})
W
wanghaoshuang 已提交
7475 7476 7477
    return out


7478
def relu(x, name=None):
W
wanghaoshuang 已提交
7479 7480
    """
    Relu takes one input data (Tensor) and produces one output data (Tensor)
7481
    where the rectified linear function, y = max(0, x), is applied to
W
wanghaoshuang 已提交
7482 7483 7484 7485
    the tensor elementwise.

    .. math::

7486
        Out = \\max(0, x)
W
wanghaoshuang 已提交
7487 7488

    Args:
7489
        x (Variable): The input tensor.
7490 7491
        name (str|None, default None): A name for this layer If set None,
            the layer will be named automatically.
W
wanghaoshuang 已提交
7492 7493 7494 7495 7496 7497 7498 7499

    Returns:
        Variable: The output tensor with the same shape as input.

    Examples:

        .. code-block:: python

7500
            output = fluid.layers.relu(x)
W
wanghaoshuang 已提交
7501 7502
    """
    helper = LayerHelper('relu', **locals())
W
wanghaoshuang 已提交
7503
    dtype = helper.input_dtype(input_param_name='x')
X
Xin Pan 已提交
7504
    out = helper.create_variable_for_type_inference(dtype)
X
Xin Pan 已提交
7505 7506
    helper.append_op(
        type="relu", inputs={"X": helper.input('x')}, outputs={"Out": out})
W
wanghaoshuang 已提交
7507
    return out
7508 7509


C
chengduo 已提交
7510 7511 7512 7513 7514 7515 7516 7517 7518 7519 7520 7521 7522 7523 7524 7525 7526 7527 7528 7529 7530 7531 7532 7533 7534 7535 7536 7537 7538 7539 7540 7541 7542 7543 7544 7545 7546 7547 7548 7549 7550
@templatedoc()
def selu(x, scale=None, alpha=None, name=None):
    """
    ${comment}

    Args:
        x (Variable): The input tensor.
        scale(float, None): If the scale is not set,
            the default value is 1.0507009873554804934193349852946.
            For more information about this value, please refer
            to: https://arxiv.org/abs/1706.02515.
        alpha(float, None): If the alpha is not set,
            the default value is 1.6732632423543772848170429916717.
            For more information about this value, please refer
            to: https://arxiv.org/abs/1706.02515.
        name (str|None, default None): A name for this layer If set None,
            the layer will be named automatically.

    Returns:
        Variable: The output tensor with the same shape as input.

    Examples:

        .. code-block:: python

            output = fluid.layers.selu(x)
    """
    helper = LayerHelper('selu', **locals())
    dtype = helper.input_dtype(input_param_name='x')
    out = helper.create_variable_for_type_inference(dtype)
    attrs = {}
    if scale is not None:
        attrs["scale"] = scale
    if alpha is not None:
        attrs["alpha"] = alpha

    helper.append_op(
        type="selu", inputs={"X": x}, outputs={"Out": out}, attrs=attrs)
    return out


W
whs 已提交
7551 7552 7553
def mean_iou(input, label, num_classes):
    """
    Mean Intersection-Over-Union is a common evaluation metric for
7554 7555 7556 7557
    semantic image segmentation, which first computes the IOU for each
    semantic class and then computes the average over classes.
    IOU is defined as follows:

W
whs 已提交
7558
    .. math::
7559

H
haowang101779990 已提交
7560
        IOU = \\frac{true\_positive}{(true\_positive + false\_positive + false\_negative)}.
W
whs 已提交
7561

7562
    The predictions are accumulated in a confusion matrix and mean-IOU
W
whs 已提交
7563 7564 7565 7566 7567
    is then calculated from it.


    Args:
        input (Variable): A Tensor of prediction results for semantic labels with type int32 or int64.
7568
        label (Variable): A Tensor of ground truth labels with type int32 or int64.
W
whs 已提交
7569
                           Its shape should be the same as input.
7570
        num_classes (int): The possible number of labels.
W
whs 已提交
7571 7572

    Returns:
M
minqiyang 已提交
7573 7574
        mean_iou (Variable),out_wrong(Variable),out_correct(Variable):

H
haowang101779990 已提交
7575
                     Three variables:
M
minqiyang 已提交
7576

H
haowang101779990 已提交
7577 7578 7579
                     - mean_iou : A Tensor representing the mean intersection-over-union with shape [1].
                     - out_wrong: A Tensor with shape [num_classes]. The wrong numbers of each class.
                     - out_correct: A Tensor with shape [num_classes]. The correct numbers of each class.
W
whs 已提交
7580 7581 7582 7583

    Examples:

        .. code-block:: python
7584

W
whs 已提交
7585 7586 7587 7588
            iou, wrongs, corrects = fluid.layers.mean_iou(predict, label, num_classes)
    """
    helper = LayerHelper('mean_iou', **locals())
    dtype = helper.input_dtype()
X
Xin Pan 已提交
7589 7590 7591
    out_mean_iou = helper.create_variable_for_type_inference(dtype='float32')
    out_wrong = helper.create_variable_for_type_inference(dtype='int32')
    out_correct = helper.create_variable_for_type_inference(dtype='int32')
W
whs 已提交
7592 7593
    helper.append_op(
        type="mean_iou",
W
whs 已提交
7594 7595
        inputs={"Predictions": input,
                "Labels": label},
W
whs 已提交
7596
        outputs={
W
whs 已提交
7597 7598 7599
            "OutMeanIou": out_mean_iou,
            "OutWrong": out_wrong,
            "OutCorrect": out_correct
W
whs 已提交
7600 7601 7602
        },
        attrs={"num_classes": num_classes})
    return out_mean_iou, out_wrong, out_correct
7603 7604 7605 7606 7607 7608 7609 7610 7611 7612 7613 7614 7615 7616 7617 7618 7619 7620 7621 7622 7623 7624 7625 7626 7627 7628 7629 7630 7631 7632 7633 7634 7635 7636 7637 7638 7639 7640 7641 7642 7643 7644 7645 7646 7647 7648 7649 7650 7651 7652 7653 7654 7655 7656 7657 7658 7659 7660 7661 7662 7663 7664 7665 7666 7667 7668 7669 7670


def crop(x, shape=None, offsets=None, name=None):
    """
    Crop input into output, as specified by offsets and shape.

    .. code-block:: text

        * Case 1:
            Given
                X = [[0, 1, 2, 0, 0]
                     [0, 3, 4, 0, 0]
                     [0, 0, 0, 0, 0]],
            and
                shape = [2, 2],
                offsets = [0, 1],
            output is:
                Out = [[1, 2],
                       [3, 4]].
        * Case 2:
            Given
                X = [[0, 1, 2, 5, 0]
                     [0, 3, 4, 6, 0]
                     [0, 0, 0, 0, 0]],
            and shape is tensor
                shape = [[0, 0, 0]
                         [0, 0, 0]]
            and
                offsets = [0, 1],

            output is:
                Out = [[1, 2, 5],
                       [3, 4, 6]].

    Args:
        x (Variable): The input tensor variable.
        shape (Variable|list/tuple of integer): The output shape is specified
            by `shape`, which can a Variable or a list/tupe of integer.
            If a tensor Variable, it's rank must be the same as `x`. This way
            is suitable for the case that the output shape may be changed each
            iteration. If a list/tupe of integer, it's length must be the same
            as the rank of `x`
        offsets (Variable|list/tuple of integer|None): Specifies the copping
            offsets at each dimension. It can be a Variable or or a list/tupe
            of integer. If a tensor Variable, it's rank must be the same as `x`.
            This way is suitable for the case that the offsets may be changed
            each iteration. If a list/tupe of integer, it's length must be the
            same as the rank of `x`. If None, the offsets are 0 at each
            dimension.
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.

    Returns:
        Variable: The cropped tensor variable.

    Raises:
        ValueError: If shape is not a list, tuple or Variable.

    Examples:

        .. code-block:: python

            x = fluid.layers.data(name="x", shape=[3, 5], dtype="float32")
            y = fluid.layers.data(name="y", shape=[2, 3], dtype="float32")
            crop = fluid.layers.crop(x, shape=y)

            # or
            z = fluid.layers.data(name="z", shape=[3, 5], dtype="float32")
T
Tink_Y 已提交
7671
            crop = fluid.layers.crop(z, shape=[-1, 2, 3])
7672 7673 7674 7675 7676

    """
    helper = LayerHelper('crop', **locals())

    if not (isinstance(shape, list) or isinstance(shape, tuple) or \
7677
            isinstance(shape, Variable)):
7678 7679 7680 7681 7682
        raise ValueError("The shape should be a list, tuple or Variable.")

    if offsets is None:
        offsets = [0] * len(x.shape)

X
Xin Pan 已提交
7683
    out = helper.create_variable_for_type_inference(x.dtype)
7684 7685 7686 7687 7688 7689 7690 7691 7692 7693 7694 7695 7696 7697 7698 7699 7700
    ipts = {'X': x}
    attrs = {}
    if isinstance(shape, Variable):
        ipts['Y'] = shape
    else:
        attrs['shape'] = shape
    if isinstance(offsets, Variable):
        ipts['Offsets'] = offsets
    else:
        attrs['offsets'] = offsets

    helper.append_op(
        type='crop',
        inputs=ipts,
        outputs={'Out': out},
        attrs=None if len(attrs) == 0 else attrs)
    return out
7701 7702


W
whs 已提交
7703 7704 7705 7706 7707 7708 7709 7710 7711 7712 7713 7714 7715 7716 7717 7718 7719
def affine_grid(theta, out_shape, name=None):
    """
    It generates a grid of (x,y) coordinates using the parameters of
    the affine transformation that correspond to a set of points where
    the input feature map should be sampled to produce the transformed
    output feature map.

    .. code-block:: text

        * Case 1:

          Given:

              theta = [[[x_11, x_12, x_13]
                        [x_14, x_15, x_16]]
                       [[x_21, x_22, x_23]
                        [x_24, x_25, x_26]]]
7720

W
whs 已提交
7721
              out_shape = [2, 3, 5, 5]
7722

W
whs 已提交
7723
          Step 1:
7724

W
whs 已提交
7725 7726 7727
              Generate normalized coordinates according to out_shape.
              The values of the normalized coordinates are in the interval between -1 and 1.
              The shape of the normalized coordinates is [2, H, W] as below:
7728

W
whs 已提交
7729 7730 7731 7732 7733 7734 7735 7736 7737 7738 7739 7740 7741 7742 7743 7744 7745 7746 7747 7748 7749 7750 7751 7752 7753 7754 7755 7756 7757 7758 7759 7760 7761 7762 7763 7764 7765 7766 7767 7768 7769 7770 7771 7772 7773
              C = [[[-1.  -1.  -1.  -1.  -1. ]
                    [-0.5 -0.5 -0.5 -0.5 -0.5]
                    [ 0.   0.   0.   0.   0. ]
                    [ 0.5  0.5  0.5  0.5  0.5]
                    [ 1.   1.   1.   1.   1. ]]
                   [[-1.  -0.5  0.   0.5  1. ]
                    [-1.  -0.5  0.   0.5  1. ]
                    [-1.  -0.5  0.   0.5  1. ]
                    [-1.  -0.5  0.   0.5  1. ]
                    [-1.  -0.5  0.   0.5  1. ]]]
              C[0] is the coordinates in height axis and  C[1] is the coordinates in width axis.

          Step2:

              Tanspose and reshape C to shape [H * W, 2] and append ones to last dimension. The we get:
              C_ = [[-1.  -1.   1. ]
                    [-0.5 -1.   1. ]
                    [ 0.  -1.   1. ]
                    [ 0.5 -1.   1. ]
                    [ 1.  -1.   1. ]
                    [-1.  -0.5  1. ]
                    [-0.5 -0.5  1. ]
                    [ 0.  -0.5  1. ]
                    [ 0.5 -0.5  1. ]
                    [ 1.  -0.5  1. ]
                    [-1.   0.   1. ]
                    [-0.5  0.   1. ]
                    [ 0.   0.   1. ]
                    [ 0.5  0.   1. ]
                    [ 1.   0.   1. ]
                    [-1.   0.5  1. ]
                    [-0.5  0.5  1. ]
                    [ 0.   0.5  1. ]
                    [ 0.5  0.5  1. ]
                    [ 1.   0.5  1. ]
                    [-1.   1.   1. ]
                    [-0.5  1.   1. ]
                    [ 0.   1.   1. ]
                    [ 0.5  1.   1. ]
                    [ 1.   1.   1. ]]
          Step3:
              Compute output by equation $$Output[i] = C_ * Theta[i]^T$$

    Args:
        theta (Variable): A batch of affine transform parameters with shape [N, 2, 3].
M
minqiyang 已提交
7774
        out_shape (Variable | list | tuple): The shape of target output with format [N, C, H, W].
H
haowang101779990 已提交
7775
                                             ``out_shape`` can be a Variable or a list or tuple.
W
whs 已提交
7776 7777 7778 7779 7780 7781 7782 7783 7784 7785 7786 7787
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.

    Returns:
        Variable: The output with shape [N, H, W, 2].

    Raises:
        ValueError: If the type of arguments is not supported.

    Examples:

        .. code-block:: python
H
haowang101779990 已提交
7788

W
whs 已提交
7789 7790 7791 7792 7793 7794 7795 7796 7797 7798 7799
            theta = fluid.layers.data(name="x", shape=[2, 3], dtype="float32")
            out_shape = fluid.layers.data(name="y", shape=[-1], dtype="float32")
            data = fluid.layers.affine_grid(theta, out_shape)

            # or
            data = fluid.layers.affine_grid(theta, [5, 3, 28, 28])

    """
    helper = LayerHelper('affine_grid')

    if not (isinstance(out_shape, list) or isinstance(out_shape, tuple) or \
7800
            isinstance(out_shape, Variable)):
W
whs 已提交
7801 7802 7803 7804 7805 7806 7807 7808 7809 7810 7811 7812 7813 7814 7815 7816 7817 7818 7819 7820 7821
        raise ValueError("The out_shape should be a list, tuple or Variable.")

    if not isinstance(theta, Variable):
        raise ValueError("The theta should be a Variable.")

    out = helper.create_variable_for_type_inference(theta.dtype)
    ipts = {'Theta': theta}
    attrs = {}
    if isinstance(out_shape, Variable):
        ipts['OutputShape'] = out_shape
    else:
        attrs['output_shape'] = out_shape

    helper.append_op(
        type='affine_grid',
        inputs=ipts,
        outputs={'Output': out},
        attrs=None if len(attrs) == 0 else attrs)
    return out


7822 7823
def rank_loss(label, left, right, name=None):
    """
H
haowang101779990 已提交
7824

7825 7826
    **Rank loss layer for RankNet**

H
haowang101779990 已提交
7827
    `RankNet <http://icml.cc/2015/wp-content/uploads/2015/06/icml_ranking.pdf>`_
7828 7829 7830
    is a pairwise ranking model with a training sample consisting of a pair
    of documents, A and B. Label P indicates whether A is ranked higher than B
    or not:
M
minqiyang 已提交
7831

7832 7833
    P = {0, 1} or {0, 0.5, 1}, where 0.5 means that there is no information
    about the rank of the input pair.
M
minqiyang 已提交
7834

H
haowang101779990 已提交
7835 7836
    Rank loss layer takes three inputs: left ( :math:`o_i` ), right ( :math:`o_j` ) and
    label ( :math:`P_{i,j}` ). The inputs respectively represent RankNet's output scores
7837 7838
    for documents A and B and the value of label P. The following equation
    computes rank loss C_{i,j} from the inputs:
M
minqiyang 已提交
7839

H
haowang101779990 已提交
7840 7841 7842 7843 7844 7845 7846 7847
    .. math::

      C_{i,j} &= -\\tilde{P_{ij}} * o_{i,j} + \log(1 + e^{o_{i,j}}) \\\\

      o_{i,j} &=  o_i - o_j  \\\\

      \\tilde{P_{i,j}} &= \\left \{0, 0.5, 1 \\right \} \ or \ \\left \{0, 1 \\right \}

M
minqiyang 已提交
7848 7849 7850

    Rank loss layer takes batch inputs with size batch_size (batch_size >= 1).

7851 7852 7853 7854 7855 7856 7857 7858 7859 7860 7861 7862 7863 7864 7865 7866 7867 7868 7869 7870 7871 7872 7873 7874 7875 7876 7877 7878 7879 7880 7881 7882 7883 7884
    Args:
        label (Variable): Indicats whether A ranked higher than B or not.
        left (Variable): RankNet's output score for doc A.
        right (Variable): RankNet's output score for doc B.
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.

    Returns:
        list: The value of rank loss.

    Raises:
        ValueError: Any of label, left, and right is not a variable.

    Examples:

        .. code-block:: python

            label = fluid.layers.data(name="label", shape=[4, 1], dtype="float32")
            left = fluid.layers.data(name="left", shape=[4, 1], dtype="float32")
            right = fluid.layers.data(name="right", shape=[4, 1], dtype="float32")
            out = fluid.layers.rank_loss(label, left, right)

    """
    helper = LayerHelper('rank_loss', **locals())

    if not (isinstance(label, Variable)):
        raise ValueError("The label should be a Variable")

    if not (isinstance(left, Variable)):
        raise ValueError("The left should be a Variable")

    if not (isinstance(right, Variable)):
        raise ValueError("The right should be a Variable")

X
Xin Pan 已提交
7885
    out = helper.create_variable_for_type_inference("float32")
7886 7887 7888 7889 7890 7891 7892 7893

    helper.append_op(
        type='rank_loss',
        inputs={"Label": label,
                "Left": left,
                "Right": right},
        outputs={'Out': out})
    return out
7894 7895


M
minqiyang 已提交
7896 7897
def margin_rank_loss(label, left, right, margin=0.1, name=None):
    """
M
minqiyang 已提交
7898
    Margin Ranking Loss Layer for ranking problem,
M
minqiyang 已提交
7899
    which compares left score and right score passed in.
M
minqiyang 已提交
7900
    The ranking loss can be defined as following equation:
M
minqiyang 已提交
7901 7902 7903

    .. math::

H
haowang101779990 已提交
7904
        rank\_loss = max(0, -label * (left - right) + margin)
M
minqiyang 已提交
7905 7906

    Args:
M
minqiyang 已提交
7907
       label (Variable): Indicates whether the left is ranked higher than the right or not.
M
minqiyang 已提交
7908 7909
       left (Variable): Ranking score for left.
       right (Variable): Ranking score for right.
M
minqiyang 已提交
7910
       margin (float): Indicates the given margin.
M
minqiyang 已提交
7911 7912
       name (str|None): A name for this layer (optional). If set None, the layer
                       will be named automatically.
H
haowang101779990 已提交
7913

M
minqiyang 已提交
7914
    Returns:
M
minqiyang 已提交
7915
       Variable: The ranking loss.
H
haowang101779990 已提交
7916

M
minqiyang 已提交
7917
    Raises:
M
minqiyang 已提交
7918
       ValueError: Any of label, left, and right is not a Variable.
H
haowang101779990 已提交
7919

M
minqiyang 已提交
7920
    Examples:
H
haowang101779990 已提交
7921

M
minqiyang 已提交
7922
        .. code-block:: python
H
haowang101779990 已提交
7923

M
minqiyang 已提交
7924 7925 7926 7927 7928
           label = fluid.layers.data(name="label", shape=[4, 1], dtype="float32")
           left = fluid.layers.data(name="left", shape=[4, 1], dtype="float32")
           right = fluid.layers.data(name="right", shape=[4, 1], dtype="float32")
           out = fluid.layers.margin_rank_loss(label, left, right)
    """
M
minqiyang 已提交
7929
    helper = LayerHelper('margin_rank_loss', **locals())
M
minqiyang 已提交
7930 7931 7932 7933 7934 7935
    if not isinstance(label, Variable):
        raise ValueError("The label should be a Variable.")
    if not isinstance(left, Variable):
        raise ValueError("The left should be a Variable.")
    if not isinstance(right, Variable):
        raise ValueError("The right should be a Variable.")
X
Xin Pan 已提交
7936 7937
    out = helper.create_variable_for_type_inference(left.dtype)
    act = helper.create_variable_for_type_inference(left.dtype)
M
minqiyang 已提交
7938 7939 7940 7941 7942 7943 7944 7945 7946 7947 7948
    helper.append_op(
        type='margin_rank_loss',
        inputs={"Label": label,
                "X1": left,
                "X2": right},
        outputs={'Out': out,
                 'Activated': act},
        attrs={'margin': margin})
    return out


W
whs 已提交
7949 7950 7951 7952 7953 7954 7955 7956 7957 7958 7959 7960
def pad2d(input,
          paddings=[0, 0, 0, 0],
          mode='constant',
          pad_value=0.0,
          data_format="NCHW",
          name=None):
    """
    Pad 2-d images accordding to 'paddings' and 'mode'.
    If mode is 'reflect', paddings[0] and paddings[1] must be no greater
    than height-1. And the width dimension has the same condition.

    Example:
T
Tink_Y 已提交
7961
        .. code-block:: text
W
whs 已提交
7962

T
Tink_Y 已提交
7963
	      Given that X is a channel of image from input:
M
minqiyang 已提交
7964

T
Tink_Y 已提交
7965 7966
	      X = [[1, 2, 3],
		   [4, 5, 6]]
M
minqiyang 已提交
7967

T
Tink_Y 已提交
7968
	      Case 0:
M
minqiyang 已提交
7969

T
Tink_Y 已提交
7970 7971 7972
		paddings = [0, 1, 2, 3],
		mode = 'constant'
		pad_value = 0
M
minqiyang 已提交
7973

T
Tink_Y 已提交
7974 7975 7976
		Out = [[0, 0, 1, 2, 3, 0, 0, 0]
		       [0, 0, 4, 5, 6, 0, 0, 0]
		       [0, 0, 0, 0, 0, 0, 0, 0]]
M
minqiyang 已提交
7977

T
Tink_Y 已提交
7978
	      Case 1:
M
minqiyang 已提交
7979

T
Tink_Y 已提交
7980 7981
		paddings = [0, 1, 2, 1],
		mode = 'reflect'
M
minqiyang 已提交
7982

T
Tink_Y 已提交
7983 7984 7985
		Out = [[3, 2, 1, 2, 3, 2]
		       [6, 5, 4, 5, 6, 5]
		       [3, 2, 1, 2, 3, 2]]
M
minqiyang 已提交
7986

T
Tink_Y 已提交
7987
	      Case 2:
M
minqiyang 已提交
7988

T
Tink_Y 已提交
7989 7990
		paddings = [0, 1, 2, 1],
		mode = 'edge'
M
minqiyang 已提交
7991

T
Tink_Y 已提交
7992 7993 7994
		Out = [[1, 1, 1, 2, 3, 3]
		       [4, 4, 4, 5, 6, 6]
		       [4, 4, 4, 5, 6, 6]]
M
minqiyang 已提交
7995 7996


W
whs 已提交
7997 7998
    Args:
        input (Variable): The input image with [N, C, H, W] format or [N, H, W, C] format.
7999
        paddings (tuple|list|Variable): The padding size. If padding is a tuple, it must
W
whs 已提交
8000 8001 8002 8003 8004 8005 8006 8007 8008 8009 8010 8011 8012 8013 8014 8015 8016 8017 8018 8019 8020 8021 8022
            contain four integers, (padding_top, padding_bottom, padding_left, padding_right).
            Default: padding = [0, 0, 0, 0].
        mode (str): Three modes: constant(default), reflect, edge. Default: constant
        pad_value (float32): The value to fill the padded areas in constant mode. Default: 0
        data_format (str): An optional string from: "NHWC", "NCHW". Specify the data format of
                           the input data.
                           Default: "NCHW"
        name (str|None): A name for this layer(optional). If set None, the layer
            will be named automatically.

    Returns:
        Variable: The tensor variable padded accordding to paddings and mode.


    Examples:
        .. code-block:: python

          data = fluid.layers.data(name='data', shape=[3, 32, 32], dtype='float32')
          result = fluid.layers.pad2d(input=data, padding=[1,2,3,4], mode='reflect')
    """

    helper = LayerHelper('pad2d', **locals())
    dtype = helper.input_dtype(input_param_name='input')
X
Xin Pan 已提交
8023
    out = helper.create_variable_for_type_inference(dtype)
8024 8025 8026 8027 8028 8029 8030 8031 8032
    inputs = {'X': input}
    attrs = {'mode': mode, 'pad_value': pad_value, 'data_format': data_format}

    if isinstance(paddings, Variable):
        inputs['Paddings'] = paddings
        attrs['paddings'] = []
    else:
        attrs['paddings'] = paddings

W
whs 已提交
8033
    helper.append_op(
8034
        type='pad2d', inputs=inputs, outputs={"Out": out}, attrs=attrs)
W
whs 已提交
8035 8036 8037 8038

    return out


8039 8040 8041 8042 8043 8044 8045 8046 8047 8048 8049 8050
@templatedoc()
def elu(x, alpha=1.0, name=None):
    """
    ${comment}
    Args:
        x(${x_type}): ${x_comment}
        alpha(${alpha_type}|1.0): ${alpha_comment}
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.

    Returns:
        output(${out_type}): ${out_comment}
Z
ZhenWang 已提交
8051 8052 8053 8054 8055

    Examples:

        .. code-block:: python

Z
ZhenWang 已提交
8056 8057
            x = fluid.layers.data(name="x", shape=[3,10,32,32], dtype="float32")
            y = fluid.layers.elu(x, alpha=0.2)
8058 8059
    """
    helper = LayerHelper('elu', **locals())
X
Xin Pan 已提交
8060
    out = helper.create_variable_for_type_inference(dtype=x.dtype)
8061 8062 8063 8064 8065 8066 8067 8068 8069 8070 8071 8072 8073 8074 8075 8076 8077 8078 8079 8080
    helper.append_op(
        type='elu',
        inputs={'X': x},
        outputs={'Out': out},
        attrs={'alpha': alpha})
    return out


@templatedoc()
def relu6(x, threshold=6.0, name=None):
    """
    ${comment}
    Args:
        x(${x_type}): ${x_comment}
        threshold(${threshold_type}|6.0): ${threshold_comment}
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.

    Returns:
        output(${out_type}): ${out_comment}
Z
ZhenWang 已提交
8081 8082 8083 8084 8085

    Examples:

        .. code-block:: python

Z
ZhenWang 已提交
8086 8087
            x = fluid.layers.data(name="x", shape=[3,10,32,32], dtype="float32")
            y = fluid.layers.relu6(x, threshold=6.0)
8088 8089
    """
    helper = LayerHelper('relu6', **locals())
X
Xin Pan 已提交
8090
    out = helper.create_variable_for_type_inference(dtype=x.dtype)
8091 8092 8093 8094 8095 8096 8097 8098 8099 8100 8101 8102 8103 8104 8105 8106 8107 8108 8109 8110
    helper.append_op(
        type='relu6',
        inputs={'X': x},
        outputs={'Out': out},
        attrs={'threshold': threshold})
    return out


@templatedoc()
def pow(x, factor=1.0, name=None):
    """
    ${comment}
    Args:
        x(${x_type}): ${x_comment}
        factor(${factor_type}|1.0): ${factor_comment}
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.

    Returns:
        output(${out_type}): ${out_comment}
Z
ZhenWang 已提交
8111 8112 8113 8114 8115

    Examples:

        .. code-block:: python

Z
ZhenWang 已提交
8116 8117
            x = fluid.layers.data(name="x", shape=[3,10,32,32], dtype="float32")
            y = fluid.layers.pow(x, factor=2.0)
8118 8119
    """
    helper = LayerHelper('pow', **locals())
X
Xin Pan 已提交
8120
    out = helper.create_variable_for_type_inference(dtype=x.dtype)
8121 8122 8123 8124 8125 8126 8127 8128 8129 8130 8131 8132 8133 8134 8135 8136 8137 8138 8139 8140 8141
    helper.append_op(
        type='pow',
        inputs={'X': x},
        outputs={'Out': out},
        attrs={'factor': factor})
    return out


@templatedoc()
def stanh(x, scale_a=2.0 / 3.0, scale_b=1.7159, name=None):
    """
    ${comment}
    Args:
        x(${x_type}): ${x_comment}
        scale_a(${scale_a_type}|2.0 / 3.0): ${scale_a_comment}
        scale_b(${scale_b_type}|1.7159): ${scale_b_comment}
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.

    Returns:
        output(${out_type}): ${out_comment}
Z
ZhenWang 已提交
8142 8143 8144 8145 8146

    Examples:

        .. code-block:: python

Z
ZhenWang 已提交
8147
            x = fluid.layers.data(name="x", shape=[3,10,32,32], dtype="float32")
Z
ZhenWang 已提交
8148
            y = fluid.layers.stanh(x, scale_a=0.67, scale_b=1.72)
8149 8150
    """
    helper = LayerHelper('stanh', **locals())
X
Xin Pan 已提交
8151
    out = helper.create_variable_for_type_inference(dtype=x.dtype)
8152 8153 8154 8155 8156 8157 8158 8159 8160 8161 8162 8163 8164 8165 8166 8167 8168 8169 8170 8171 8172 8173
    helper.append_op(
        type='stanh',
        inputs={'X': x},
        outputs={'Out': out},
        attrs={'scale_a': scale_a,
               'scale_b': scale_b})
    return out


@templatedoc()
def hard_sigmoid(x, slope=0.2, offset=0.5, name=None):
    """
    ${comment}
    Args:
        x(${x_type}): ${x_comment}
        slope(${slope_type}|0.2): ${slope_comment}
        offset(${offset_type}|0.5): ${offset_comment}
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.

    Returns:
        output(${out_type}): ${out_comment}
Z
ZhenWang 已提交
8174 8175 8176 8177 8178

    Examples:

        .. code-block:: python

Z
ZhenWang 已提交
8179 8180
            x = fluid.layers.data(name="x", shape=[3,10,32,32], dtype="float32")
            y = fluid.layers.hard_sigmoid(x, slope=0.3, offset=0.8)
8181 8182
    """
    helper = LayerHelper('hard_sigmoid', **locals())
X
Xin Pan 已提交
8183
    out = helper.create_variable_for_type_inference(dtype=x.dtype)
8184 8185 8186 8187 8188 8189 8190 8191 8192 8193 8194 8195 8196 8197 8198 8199 8200 8201 8202 8203 8204
    helper.append_op(
        type='hard_sigmoid',
        inputs={'X': x},
        outputs={'Out': out},
        attrs={'slope': slope,
               'offset': offset})
    return out


@templatedoc()
def swish(x, beta=1.0, name=None):
    """
    ${comment}
    Args:
        x(${x_type}): ${x_comment}
        beta(${beta_type}|1.0): ${beta_comment}
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.

    Returns:
        output(${out_type}): ${out_comment}
Z
ZhenWang 已提交
8205 8206 8207 8208 8209

    Examples:

        .. code-block:: python

Z
ZhenWang 已提交
8210 8211
            x = fluid.layers.data(name="x", shape=[3,10,32,32], dtype="float32")
            y = fluid.layers.swish(x, beta=2.0)
8212 8213
    """
    helper = LayerHelper('swish', **locals())
X
Xin Pan 已提交
8214
    out = helper.create_variable_for_type_inference(dtype=x.dtype)
8215 8216 8217 8218 8219 8220 8221 8222
    helper.append_op(
        type='swish',
        inputs={'X': x},
        outputs={'Out': out},
        attrs={'slope': beta})
    return out


J
jerrywgz 已提交
8223 8224 8225 8226
def prelu(x, mode, param_attr=None, name=None):
    """
    Equation:

H
haowang101779990 已提交
8227 8228
    .. math::
        y = \max(0, x) + \\alpha * \min(0, x)
J
jerrywgz 已提交
8229 8230 8231

    Args:
        x (Variable): The input tensor.
J
jerrywgz 已提交
8232
        param_attr(ParamAttr|None): The parameter attribute for the learnable
T
Tink_Y 已提交
8233
          weight (alpha).
J
jerrywgz 已提交
8234
        mode (string): The mode for weight sharing. It supports all, channel
T
Tink_Y 已提交
8235 8236 8237
          and element. all: all elements share same weight
          channel:elements in a channel share same weight
          element:each element has a weight
J
jerrywgz 已提交
8238
        name(str|None): A name for this layer(optional). If set None, the layer
T
Tink_Y 已提交
8239
          will be named automatically.
J
jerrywgz 已提交
8240 8241 8242 8243 8244 8245 8246 8247

    Returns:
        Variable: The output tensor with the same shape as input.

    Examples:

        .. code-block:: python

J
jerrywgz 已提交
8248
            x = fluid.layers.data(name="x", shape=[10,10], dtype="float32")
J
jerrywgz 已提交
8249 8250 8251 8252 8253 8254 8255 8256 8257 8258 8259 8260 8261
            mode = 'channel'
            output = fluid.layers.prelu(x,mode)
    """
    helper = LayerHelper('prelu', **locals())
    if mode not in ['all', 'channel', 'element']:
        raise ValueError('mode should be one of all, channel, element.')
    alpha_shape = [1]
    if mode == 'channel':
        alpha_shape = [1, x.shape[1], 1, 1]
    elif mode == 'element':
        alpha_shape = x.shape
    dtype = helper.input_dtype(input_param_name='x')
    alpha = helper.create_parameter(
Q
Qiao Longfei 已提交
8262
        attr=helper.param_attr,
J
jerrywgz 已提交
8263 8264 8265 8266
        shape=alpha_shape,
        dtype='float32',
        is_bias=False,
        default_initializer=Constant(1.0))
X
Xin Pan 已提交
8267
    out = helper.create_variable_for_type_inference(dtype)
J
jerrywgz 已提交
8268 8269 8270 8271 8272 8273 8274 8275 8276
    helper.append_op(
        type="prelu",
        inputs={"X": x,
                'Alpha': alpha},
        attrs={"mode": mode},
        outputs={"Out": out})
    return out


8277 8278 8279 8280 8281 8282 8283 8284 8285 8286
@templatedoc()
def brelu(x, t_min=0.0, t_max=24.0, name=None):
    """
    ${comment}
    Args:
        x(${x_type}): ${x_comment}
        t_min(${t_min_type}|0.0): ${t_min_comment}
        t_max(${t_max_type}|24.0): ${t_max_comment}
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.
8287
    Returns:
8288
        output(${out_type}): ${out_comment}
8289 8290 8291

    Examples:

8292
    .. code-block:: python
8293

H
haowang101779990 已提交
8294 8295
            x = fluid.layers.data(name="x", shape=[2,3,16,16], dtype="float32")
            y = fluid.layers.brelu(x, t_min=1.0, t_max=20.0)
8296 8297
    """
    helper = LayerHelper('brelu', **locals())
X
Xin Pan 已提交
8298
    out = helper.create_variable_for_type_inference(dtype=x.dtype)
8299 8300 8301 8302 8303 8304 8305 8306 8307 8308 8309 8310 8311 8312 8313 8314 8315 8316
    helper.append_op(
        type='brelu',
        inputs={'X': x},
        outputs={'Out': out},
        attrs={'t_min': t_min,
               't_max': t_max})
    return out


@templatedoc()
def leaky_relu(x, alpha=0.02, name=None):
    """
    ${comment}
    Args:
        x(${x_type}): ${x_comment}
        alpha(${alpha_type}|0.02): ${alpha_comment}
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.
8317
    Returns:
8318
        output(${out_type}): ${out_comment}
8319 8320 8321 8322 8323

    Examples:

        .. code-block:: python

H
haowang101779990 已提交
8324 8325
            x = fluid.layers.data(name="x", shape=[2,3,16,16], dtype="float32")
            y = fluid.layers.leaky_relu(x, alpha=0.01)
8326 8327
    """
    helper = LayerHelper('leaky_relu', **locals())
X
Xin Pan 已提交
8328
    out = helper.create_variable_for_type_inference(dtype=x.dtype)
8329 8330 8331 8332 8333 8334 8335 8336 8337 8338 8339 8340 8341 8342 8343 8344 8345
    helper.append_op(
        type='leaky_relu',
        inputs={'X': x},
        outputs={'Out': out},
        attrs={'alpha': alpha})
    return out


@templatedoc()
def soft_relu(x, threshold=40.0, name=None):
    """
    ${comment}
    Args:
        x(${x_type}): ${x_comment}
        threshold(${threshold_type}|40.0): ${threshold_comment}
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.
8346
    Returns:
8347
        output(${out_type}): ${out_comment}
8348 8349 8350 8351 8352

    Examples:

        .. code-block:: python

H
haowang101779990 已提交
8353 8354
            x = fluid.layers.data(name="x", shape=[2,3,16,16], dtype="float32")
            y = fluid.layers.soft_relu(x, threshold=20.0)
8355 8356
    """
    helper = LayerHelper('soft_relu', **locals())
X
Xin Pan 已提交
8357
    out = helper.create_variable_for_type_inference(dtype=x.dtype)
8358 8359 8360 8361 8362 8363 8364 8365
    helper.append_op(
        type='soft_relu',
        inputs={'X': x},
        outputs={'Out': out},
        attrs={'threshold': threshold})
    return out


8366 8367 8368 8369
def flatten(x, axis=1, name=None):
    """
    **Flatten layer**
    Flattens the input tensor into a 2D matrix.
M
minqiyang 已提交
8370

H
haowang101779990 已提交
8371
    For Example:
M
minqiyang 已提交
8372

H
haowang101779990 已提交
8373
    .. code-block:: text
8374

H
haowang101779990 已提交
8375 8376 8377 8378 8379 8380 8381 8382 8383 8384 8385 8386 8387 8388 8389 8390 8391 8392 8393 8394 8395
        Case 1:

          Given
            X.shape = (3, 100, 100, 4)

          and
            axis = 2

          We get:
            Out.shape = (3 * 100, 4 * 100)

        Case 2:

          Given
            X.shape = (3, 100, 100, 4)

          and
            axis = 0

          We get:
            Out.shape = (1, 3 * 100 * 100 * 4)
8396 8397 8398

    Args:
        x (Variable): A tensor of rank >= axis.
8399 8400
        axis (int): Indicate up to which input dimensions (exclusive) should
                    be flattened to the outer dimension of the output.
8401 8402 8403 8404 8405 8406 8407 8408
                    The value for axis must be in the range [0, R], where R
                    is the rank of the input tensor. When axis = 0, the shape
                    of the output tensor is (1, (d_0 X d_1 ... d_n), where the
                    shape of the input tensor is (d_0, d_1, ... d_n).
        name(str|None): A name for this layer(optional). If set None, the layer
                        will be named automatically.

    Returns:
H
haowang101779990 已提交
8409 8410 8411
        Variable: A 2D tensor with the contents of the input tensor, with input \
                  dimensions up to axis flattened to the outer dimension of \
                  the output and remaining input dimensions flattened into the \
8412 8413 8414 8415
                  inner dimension of the output.

    Raises:
        ValueError: If x is not a variable.
8416
        ValueError: If axis is not in range [0, rank(x)].
8417 8418 8419 8420 8421 8422 8423 8424 8425 8426 8427 8428 8429 8430 8431 8432

    Examples:

        .. code-block:: python

            x = fluid.layers.data(name="x", shape=[4, 4, 3], dtype="float32")
            out = fluid.layers.flatten(x=x, axis=2)
    """
    helper = LayerHelper('flatten', **locals())

    if not (isinstance(x, Variable)):
        raise ValueError("The input x should be a Variable")

    if not (isinstance(axis, int)) or axis > len(x.shape) or axis < 0:
        raise ValueError("The axis should be a int, and in range [0, rank(x)]")

X
Xin Pan 已提交
8433 8434
    out = helper.create_variable_for_type_inference(x.dtype)
    x_shape = helper.create_variable_for_type_inference(x.dtype)
8435
    helper.append_op(
8436
        type='flatten2',
8437
        inputs={"X": x},
8438 8439
        outputs={'Out': out,
                 'XShape': x_shape},
8440 8441
        attrs={"axis": axis})
    return out
X
Xin Pan 已提交
8442 8443


C
chenweihang 已提交
8444
def sequence_enumerate(input, win_size, pad_value=0, name=None):
C
chenweihang 已提交
8445
    """
C
chenweihang 已提交
8446
    Generate a new sequence for the input index sequence, which enumerates all the
M
minqiyang 已提交
8447
    sub-sequences with length `win_size` of the input.
C
chenweihang 已提交
8448 8449
    The enumerated sequence has the same 1st dimension with variable `input`, and
    the 2nd dimension is `win_size`, padded by `pad_value` if necessary in generation.
M
minqiyang 已提交
8450

H
haowang101779990 已提交
8451 8452 8453 8454 8455 8456 8457 8458 8459 8460 8461 8462 8463 8464 8465 8466 8467
    .. code-block:: text

        Case 1:

          Input:
            X.lod = [[0, 3, 5]]
            X.data = [[1], [2], [3], [4], [5]]
            X.dims = [5, 1]

          Attrs:
            win_size = 2
            pad_value = 0

          Output:
            Out.lod = [[0, 3, 5]]
            Out.data = [[1, 2], [2, 3], [3, 0], [4, 5], [5, 0]]
            Out.dims = [5, 2]
C
chenweihang 已提交
8468 8469

    Args:
C
chenweihang 已提交
8470 8471 8472
        input (Variable): The input variable which is a index sequence.
        win_size (int): The window size for enumerating all sub-sequences.
        pad_value (int): The padding value, default 0.
C
chenweihang 已提交
8473 8474 8475 8476 8477 8478 8479 8480 8481 8482 8483

    Returns:
        Variable: The enumerate sequence variable which is a LoDTensor.

    Examples:
        .. code-block:: python

            x = fluid.layers.data(shape[30, 1], dtype='int32', lod_level=1)
            out = fluid.layers.sequence_enumerate(input=x, win_size=3, pad_value=0)
    """
    helper = LayerHelper('sequence_enumerate', **locals())
X
Xin Pan 已提交
8484 8485
    out = helper.create_variable_for_type_inference(
        helper.input_dtype(), stop_gradient=True)
C
chenweihang 已提交
8486 8487 8488 8489 8490 8491
    helper.append_op(
        type='sequence_enumerate',
        inputs={'X': input},
        outputs={'Out': out},
        attrs={'win_size': win_size,
               'pad_value': pad_value})
M
minqiyang 已提交
8492
    return out
8493

8494

S
sneaxiy 已提交
8495 8496 8497 8498 8499 8500 8501 8502 8503
def sequence_mask(x, maxlen=None, dtype='int64', name=None):
    """
    **SequenceMask Layer**

    This layer outputs a mask according to the input :code:`x` and
    :code:`maxlen` with data type of :code:`dtype`.

    Supposing :code:`x` is a Tensor with shape [d_1, d_2, ..., d_n], the
    :code:`y` is a mask with shape [d_1, d_2, ..., d_n, maxlen], where:
8504

S
sneaxiy 已提交
8505
    .. math::
8506

S
sneaxiy 已提交
8507 8508 8509
        y(i_1, i_2,..., i_n, j) = (j < x(i_1, i_2,..., i_n))

    Args:
8510
        x (Variable): Input tensor of sequence_mask layer,
S
sneaxiy 已提交
8511 8512 8513 8514
                      whose elements are integers less than :code:`maxlen`.
        maxlen (int|None): Maximum length of the sequence. If :code:`maxlen`
                           is None, it would be replace with :math:`max(x)`.
        dtype (np.dtype|core.VarDesc.VarType|str): Data type of the output.
8515 8516 8517
        name (str|None): A name for this layer(optional). If set None, the
                         layer will be named automatically.

S
sneaxiy 已提交
8518 8519
    Returns:
        Variable: The output sequence mask.
8520

S
sneaxiy 已提交
8521 8522
    """

Q
qingqing01 已提交
8523
    helper = LayerHelper('sequence_mask', **locals())
S
sneaxiy 已提交
8524
    if name is None:
X
Xin Pan 已提交
8525
        out = helper.create_variable_for_type_inference(dtype=dtype)
S
sneaxiy 已提交
8526
    else:
X
Xin Pan 已提交
8527
        out = helper.create_variable_for_type_inference(dtype=dtype, name=name)
S
sneaxiy 已提交
8528

Q
qingqing01 已提交
8529 8530 8531
    helper.append_op(
        type='sequence_mask',
        inputs={'X': [x]},
S
sneaxiy 已提交
8532 8533
        outputs={'Y': out},
        attrs={
8534
            'maxlen': maxlen if maxlen is not None else -1,
S
sneaxiy 已提交
8535 8536 8537
            'out_dtype': out.dtype
        })
    return out
S
sneaxiy 已提交
8538 8539


X
Xin Pan 已提交
8540
def stack(x, axis=0):
S
sneaxiy 已提交
8541 8542 8543 8544
    """
    **Stack Layer**

    This layer stacks all of the input :code:`x` along axis.
8545 8546 8547 8548 8549 8550 8551

    Input :code:`x` can be a single variable, a :code:`list` of variables,
    or a :code:`tuple` of variables. If :code:`x` is a :code:`list` or
    :code:`tuple`, the shapes of all these variables must be the same.
    Supposing the shape of each input is :math:`[d_0, d_1, ..., d_{n-1}]`,
    the shape of the output variable would be
    :math:`[d_0, d_1, ..., d_{axis}=len(x), ..., d_{n-1}]`.
S
sneaxiy 已提交
8552
    If :code:`axis` < 0, it would be replaced with :code:`axis+rank(x[0])+1`.
8553
    If :code:`axis` is None, it would be replaced with 0.
S
sneaxiy 已提交
8554

C
chengduozh 已提交
8555 8556
    For Example:

C
chengduozh 已提交
8557 8558 8559 8560 8561 8562 8563 8564 8565 8566 8567 8568 8569 8570 8571 8572 8573 8574 8575 8576 8577 8578 8579 8580 8581 8582 8583 8584 8585 8586 8587 8588 8589 8590 8591 8592 8593 8594
    .. code-block:: text

        Case 1:
          Input:
            x[0].data = [ [1.0 , 2.0 ] ]
            x[0].dims = [1, 2]
            x[1].data = [ [3.0 , 4.0 ] ]
            x[1].dims = [1, 2]
            x[2].data = [ [5.0 , 6.0 ] ]
            x[2].dims = [1, 2]

          Attrs:
            axis = 0

          Output:
            Out.data =[ [ [1.0, 2.0] ],
                        [ [3.0, 4.0] ],
                        [ [5.0, 6.0] ] ]
            Out.dims = [3, 1, 2]

        Case 2:
          Given
            x[0].data = [ [1.0 , 2.0 ] ]
            x[0].dims = [1, 2]
            x[1].data = [ [3.0 , 4.0 ] ]
            x[1].dims = [1, 2]
            x[2].data = [ [5.0 , 6.0 ] ]
            x[2].dims = [1, 2]

          Attrs:
            axis = 1 or axis = -2

          Output:
            Out.data =[ [ [1.0, 2.0]
                          [3.0, 4.0]
                          [5.0, 6.0] ] ]
            Out.dims = [1, 3, 2]

S
sneaxiy 已提交
8595
    Args:
8596
        x (Variable|list(Variable)|tuple(Variable)): Input variables.
S
sneaxiy 已提交
8597
        axis (int|None): The axis along which all inputs are stacked.
8598

S
sneaxiy 已提交
8599 8600
    Returns:
        Variable: The stacked variable.
8601

S
sneaxiy 已提交
8602 8603
    """

X
Xin Pan 已提交
8604 8605 8606 8607 8608 8609
    helper = LayerHelper('stack', **locals())
    axis = 0 if axis is None else axis

    if not isinstance(x, list) and not isinstance(x, tuple):
        x = [x]

X
Xin Pan 已提交
8610
    out = helper.create_variable_for_type_inference(x[0].dtype)
X
Xin Pan 已提交
8611
    helper.append_op(
S
sneaxiy 已提交
8612 8613
        type='stack', inputs={'X': x}, outputs={'Y': out},
        attrs={'axis': axis})
8614

X
Xin Pan 已提交
8615
    return out
D
dzhwinter 已提交
8616 8617 8618 8619 8620 8621 8622


def unstack(x, axis=0, num=None):
    """
    **UnStack Layer**

    This layer unstacks input :code:`x` into several tensors along axis.
M
minqiyang 已提交
8623

D
dzhwinter 已提交
8624 8625 8626
    If :code:`axis` < 0, it would be replaced with :code:`axis+rank(x)`.
    If :code:`num` is None, it would be inferred from :code:`x.shape[axis]`,
    and if :code:`x.shape[axis]` <= 0 or is unknown, :code:`ValueError` is
M
minqiyang 已提交
8627
    raised.
D
dzhwinter 已提交
8628 8629

    Args:
M
minqiyang 已提交
8630
        x (Variable): Input variable.
D
dzhwinter 已提交
8631 8632
        axis (int): The axis along which the input is unstacked.
        num (int|None): The number of output variables.
M
minqiyang 已提交
8633

D
dzhwinter 已提交
8634 8635
    Returns:
        list(Variable): The unstacked variables.
M
minqiyang 已提交
8636

D
dzhwinter 已提交
8637 8638 8639 8640 8641 8642 8643 8644 8645 8646
    """

    helper = LayerHelper('unstack', **locals())
    if num is None:
        if axis is None or x.shape[axis] <= 0:
            raise ValueError('unknown unstack number')
        else:
            num = x.shape[axis]

    outs = []
Y
Yibing Liu 已提交
8647
    for _ in range(num):
X
Xin Pan 已提交
8648
        outs.append(helper.create_variable_for_type_inference(x.dtype))
D
dzhwinter 已提交
8649 8650 8651 8652 8653 8654 8655 8656

    helper.append_op(
        type='unstack',
        inputs={'X': [x]},
        outputs={'Y': outs},
        attrs={'axis': axis,
               'num': num})
    return outs
W
whs 已提交
8657 8658 8659 8660 8661 8662 8663 8664 8665 8666 8667 8668


def expand(x, expand_times, name=None):
    """Expand operator tiles the input by given times number. You should set times
    number for each dimension by providing attribute 'expand_times'. The rank of X
    should be in [1, 6]. Please note that size of 'expand_times' must be the same
    with X's rank. Following is a using case:


    .. code-block:: text

        Input(X) is a 3-D tensor with shape [2, 3, 1]:
M
minqiyang 已提交
8669

W
whs 已提交
8670 8671 8672 8673
                [
                   [[1], [2], [3]],
                   [[4], [5], [6]]
                ]
M
minqiyang 已提交
8674

W
whs 已提交
8675
        Attr(expand_times):  [1, 2, 2]
M
minqiyang 已提交
8676

W
whs 已提交
8677
        Output(Out) is a 3-D tensor with shape [2, 6, 2]:
M
minqiyang 已提交
8678

W
whs 已提交
8679 8680 8681 8682
                [
                    [[1, 1], [2, 2], [3, 3], [1, 1], [2, 2], [3, 3]],
                    [[4, 4], [5, 5], [6, 6], [4, 4], [5, 5], [6, 6]]
                ]
M
minqiyang 已提交
8683

W
whs 已提交
8684 8685 8686 8687 8688 8689 8690 8691 8692 8693 8694 8695 8696 8697 8698 8699
    Args:
        x (Variable): A tensor with rank in [1, 6].
        expand_times (list|tuple): Expand times number for each dimension.

    Returns:
        Variable: The expanded variable which is a LoDTensor. After expanding, size of each dimension of Output(Out) is equal to ithe size of the corresponding dimension of Input(X) multiplying the corresponding value given by expand_times.


    Examples:
        .. code-block:: python

            x = fluid.layers.data(name='x', shape=[10], dtype='float32')
            out = fluid.layers.expand(x=x, expand_times=[1, 2, 2])
    """
    helper = LayerHelper('expand', input=x, **locals())
    dtype = helper.input_dtype(input_param_name='x')
X
Xin Pan 已提交
8700
    out = helper.create_variable_for_type_inference(dtype)
W
whs 已提交
8701 8702 8703 8704 8705 8706
    helper.append_op(
        type='expand',
        inputs={'X': x},
        outputs={'Out': out},
        attrs={'expand_times': expand_times})
    return out
S
sneaxiy 已提交
8707 8708


G
fix  
gongweibao 已提交
8709 8710 8711
from paddle.fluid.framework import convert_np_dtype_to_dtype_


G
gongweibao 已提交
8712
@templatedoc()
G
fix  
gongweibao 已提交
8713 8714 8715 8716 8717 8718 8719 8720 8721
def uniform_random_batch_size_like(input,
                                   shape,
                                   dtype='float32',
                                   input_dim_idx=0,
                                   output_dim_idx=0,
                                   min=-1.0,
                                   max=1.0,
                                   seed=0):
    """
G
gongweibao 已提交
8722
    ${comment}
G
fix  
gongweibao 已提交
8723 8724

    Args:
G
gongweibao 已提交
8725 8726 8727
        input (Variable): ${input_comment}
        shape (tuple|list): ${shape_comment}
        input_dim_idx (Int): ${input_dim_idx_comment}
G
gongweibao 已提交
8728
        output_dim_idx (Int): ${output_dim_idx_comment}
G
gongweibao 已提交
8729 8730 8731
        min (Float): ${min_comment}
        max (Float): ${max_comment}
        seed (Int): ${seed_comment}
G
fix  
gongweibao 已提交
8732 8733
        dtype(np.dtype|core.VarDesc.VarType|str): The type of data : float32, float_16, int etc
    Returns:
G
gongweibao 已提交
8734
        out (Variable): ${out_comment}
G
fix  
gongweibao 已提交
8735

8736 8737 8738 8739 8740
    Examples:
        .. code-block:: python

            input = layers.data(name="input", shape=[13, 11], dtype='float32')
            out = layers.uniform_random_batch_size_like(input, [-1, 11])
G
fix  
gongweibao 已提交
8741 8742 8743
    """

    helper = LayerHelper('uniform_random_batch_size_like', **locals())
X
Xin Pan 已提交
8744
    out = helper.create_variable_for_type_inference(dtype)
G
fix  
gongweibao 已提交
8745 8746 8747 8748 8749 8750 8751 8752 8753 8754 8755 8756 8757 8758 8759 8760
    c_dtype = convert_np_dtype_to_dtype_(dtype)
    helper.append_op(
        type='uniform_random_batch_size_like',
        inputs={'Input': input},
        outputs={'Out': out},
        attrs={
            'shape': shape,
            'input_dim_idx': input_dim_idx,
            'output_dim_idx': output_dim_idx,
            'min': min,
            'max': max,
            'seed': seed,
            'dtype': c_dtype
        })

    return out
G
fix  
gongweibao 已提交
8761 8762


G
gongweibao 已提交
8763
@templatedoc()
X
Xin Pan 已提交
8764
def gaussian_random(shape, mean=0.0, std=1.0, seed=0, dtype='float32'):
G
fix  
gongweibao 已提交
8765
    """
G
gongweibao 已提交
8766
    ${comment}
G
fix  
gongweibao 已提交
8767 8768

    Args:
G
gongweibao 已提交
8769 8770 8771 8772
        shape (tuple|list): ${shape_comment}
        mean (Float): ${mean_comment}
        std (Float): ${std_comment}
        seed (Int): ${seed_comment}
G
fix  
gongweibao 已提交
8773 8774 8775
        dtype(np.dtype|core.VarDesc.VarType|str): Output data type.

    Returns:
G
gongweibao 已提交
8776
        out (Variable): ${out_comment}
G
fix  
gongweibao 已提交
8777

8778 8779 8780 8781
    Examples:
        .. code-block:: python

            out = layers.gaussian_random(shape=[20, 30])
G
fix  
gongweibao 已提交
8782 8783 8784
    """

    helper = LayerHelper('gaussian_random', **locals())
X
Xin Pan 已提交
8785
    out = helper.create_variable_for_type_inference(dtype)
G
fix  
gongweibao 已提交
8786 8787 8788 8789 8790 8791 8792 8793 8794 8795
    c_dtype = convert_np_dtype_to_dtype_(dtype)
    helper.append_op(
        type='gaussian_random',
        outputs={'Out': out},
        attrs={
            'shape': shape,
            'mean': mean,
            'std': std,
            'seed': seed,
            'dtype': c_dtype,
X
Xin Pan 已提交
8796
            'use_mkldnn': False
G
fix  
gongweibao 已提交
8797 8798 8799 8800 8801
        })

    return out


G
gongweibao 已提交
8802
@templatedoc()
G
fix  
gongweibao 已提交
8803
def sampling_id(x, min=0.0, max=1.0, seed=0, dtype='float32'):
G
fix  
gongweibao 已提交
8804
    """
G
gongweibao 已提交
8805
    ${comment}
G
fix  
gongweibao 已提交
8806 8807

    Args:
G
gongweibao 已提交
8808 8809 8810 8811
        x (Variable): ${x_comment}
        min (Float): ${min_comment}
        max (Float): ${max_comment}
        seed (Float): ${seed_comment}
G
fix  
gongweibao 已提交
8812
        dtype(np.dtype|core.VarDesc.VarType|str): The type of output data : float32, float_16, int etc
G
fix  
gongweibao 已提交
8813 8814

    Returns:
G
gongweibao 已提交
8815
        out (Variable): ${out_comment}
G
fix  
gongweibao 已提交
8816

8817 8818 8819 8820 8821 8822 8823 8824 8825 8826
    Examples:
        .. code-block:: python

            x = layers.data(
                name="X",
                shape=[13, 11],
                dtype='float32',
                append_batch_size=False)

            out = layers.sampling_id(x)
G
fix  
gongweibao 已提交
8827 8828 8829
    """

    helper = LayerHelper('sampling_id', **locals())
X
Xin Pan 已提交
8830
    out = helper.create_variable_for_type_inference(dtype)
G
fix  
gongweibao 已提交
8831 8832 8833 8834 8835 8836 8837 8838 8839 8840 8841
    helper.append_op(
        type='sampling_id',
        inputs={'X': x},
        outputs={'Out': out},
        attrs={'min': min,
               'max': max,
               'seed': seed})

    return out


G
gongweibao 已提交
8842
@templatedoc()
G
fix  
gongweibao 已提交
8843 8844 8845 8846 8847 8848 8849 8850 8851
def gaussian_random_batch_size_like(input,
                                    shape,
                                    input_dim_idx=0,
                                    output_dim_idx=0,
                                    mean=0.0,
                                    std=1.0,
                                    seed=0,
                                    dtype='float32'):
    """
G
gongweibao 已提交
8852
    ${comment}
G
fix  
gongweibao 已提交
8853 8854

    Args:
G
gongweibao 已提交
8855 8856
        input (Variable): ${input_comment}
        shape (tuple|list): ${shape_comment}
G
gongweibao 已提交
8857
        input_dim_idx (Int): ${input_dim_idx_comment}
G
gongweibao 已提交
8858 8859 8860 8861
        output_dim_idx (Int): ${output_dim_idx_comment}
        mean (Float): ${mean_comment}
        std (Float): ${std_comment}
        seed (Int): ${seed_comment}
G
fix  
gongweibao 已提交
8862
        dtype(np.dtype|core.VarDesc.VarType|str): The type of output data : float32, float_16, int etc
G
fix  
gongweibao 已提交
8863 8864

    Returns:
G
gongweibao 已提交
8865
        out (Variable): ${out_comment}
8866 8867 8868 8869 8870 8871 8872 8873

    Examples:
        .. code-block:: python

            input = layers.data(name="input", shape=[13, 11], dtype='float32')

            out = layers.gaussian_random_batch_size_like(
                input, shape=[-1, 11], mean=1.0, std=2.0)
G
fix  
gongweibao 已提交
8874 8875 8876
    """

    helper = LayerHelper('gaussian_random_batch_size_like', **locals())
X
Xin Pan 已提交
8877
    out = helper.create_variable_for_type_inference(dtype)
G
fix  
gongweibao 已提交
8878 8879 8880 8881 8882 8883 8884 8885 8886 8887 8888 8889 8890 8891 8892 8893 8894 8895
    c_dtype = convert_np_dtype_to_dtype_(dtype)
    helper.append_op(
        type='gaussian_random_batch_size_like',
        inputs={'Input': input},
        outputs={'Out': out},
        attrs={
            'shape': shape,
            'input_dim_idx': input_dim_idx,
            'output_dim_idx': output_dim_idx,
            'mean': mean,
            'std': std,
            'seed': seed,
            'dtype': c_dtype
        })

    return out


G
gongweibao 已提交
8896
@templatedoc()
X
Xin Pan 已提交
8897
def sum(x):
G
fix  
gongweibao 已提交
8898
    """
G
gongweibao 已提交
8899
    ${comment}
G
fix  
gongweibao 已提交
8900 8901

    Args:
G
gongweibao 已提交
8902
        x (Variable): ${x_comment}
G
fix  
gongweibao 已提交
8903 8904

    Returns:
G
gongweibao 已提交
8905
        out (Variable): ${out_comment}
8906 8907 8908 8909 8910 8911

    Examples:
        .. code-block:: python

            input = layers.data(name="input", shape=[13, 11], dtype='float32')
            out = layers.sum(input)
G
fix  
gongweibao 已提交
8912 8913 8914
    """

    helper = LayerHelper('sum', **locals())
X
Xin Pan 已提交
8915 8916
    out = helper.create_variable_for_type_inference(
        dtype=helper.input_dtype('x'))
G
fix  
gongweibao 已提交
8917 8918 8919 8920
    helper.append_op(
        type='sum',
        inputs={'X': x},
        outputs={'Out': out},
X
Xin Pan 已提交
8921
        attrs={'use_mkldnn': False})
G
fix  
gongweibao 已提交
8922 8923 8924 8925

    return out


G
gongweibao 已提交
8926
@templatedoc()
G
fix  
gongweibao 已提交
8927 8928
def slice(input, axes, starts, ends):
    """
G
gongweibao 已提交
8929
    ${comment}
G
fix  
gongweibao 已提交
8930 8931

    Args:
G
gongweibao 已提交
8932 8933 8934 8935
        input (Variable): ${input_comment}.
        axes (List): ${axes_comment}
        starts (List): ${starts_comment}
        ends (List): ${ends_comment}
G
fix  
gongweibao 已提交
8936 8937

    Returns:
G
gongweibao 已提交
8938
        out (Variable): ${out_comment}
G
fix  
gongweibao 已提交
8939

8940 8941 8942 8943 8944 8945 8946 8947 8948 8949 8950
    Examples:
        .. code-block:: python

            starts = [1, 0, 2]
            ends = [3, 3, 4]
            axes = [0, 1, 2]

            input = layers.data(
                name="input", shape=[3, 4, 5, 6], dtype='float32')

            out = layers.slice(input, axes=axes, starts=starts, ends=ends)
G
fix  
gongweibao 已提交
8951 8952 8953
    """

    helper = LayerHelper('slice', **locals())
X
Xin Pan 已提交
8954 8955
    out = helper.create_variable_for_type_inference(
        dtype=helper.input_dtype('input'))
G
fix  
gongweibao 已提交
8956 8957 8958 8959 8960 8961 8962 8963 8964 8965 8966 8967 8968
    helper.append_op(
        type='slice',
        inputs={'Input': input},
        outputs={'Out': out},
        attrs={'axes': axes,
               'starts': starts,
               'ends': ends})

    return out


def shape(input):
    """
C
chengduozh 已提交
8969 8970
    **Shape Layer**

C
fix doc  
chengduozh 已提交
8971
    Get the shape of the input.
G
fix  
gongweibao 已提交
8972 8973

    Args:
C
chengduozh 已提交
8974
        input (Variable): The input variable.
G
fix  
gongweibao 已提交
8975 8976

    Returns:
C
fix doc  
chengduozh 已提交
8977
        Variable: The shape of the input variable.
G
fix  
gongweibao 已提交
8978

8979 8980 8981 8982 8983 8984
    Examples:
        .. code-block:: python

            input = layers.data(
                name="input", shape=[3, 100, 100], dtype="float32")
            out = layers.shape(input)
G
fix  
gongweibao 已提交
8985 8986 8987
    """

    helper = LayerHelper('shape', **locals())
8988
    out = helper.create_variable_for_type_inference(dtype='int32')
G
fix  
gongweibao 已提交
8989
    helper.append_op(
G
fix  
gongweibao 已提交
8990
        type='shape', inputs={'Input': input}, outputs={'Out': out})
G
fix  
gongweibao 已提交
8991 8992

    return out
G
merge  
gongweibao 已提交
8993 8994


S
sneaxiy 已提交
8995 8996 8997 8998 8999 9000 9001 9002
def _elementwise_op(helper):
    op_type = helper.layer_type
    x = helper.kwargs.get('x', None)
    y = helper.kwargs.get('y', None)
    assert x is not None, 'x cannot be None in {}'.format(op_type)
    assert y is not None, 'y cannot be None in {}'.format(op_type)
    axis = helper.kwargs.get('axis', -1)
    use_mkldnn = helper.kwargs.get('use_mkldnn', False)
S
sneaxiy 已提交
9003 9004
    name = helper.kwargs.get('name', None)
    if name is None:
X
Xin Pan 已提交
9005
        out = helper.create_variable_for_type_inference(dtype=x.dtype)
S
sneaxiy 已提交
9006 9007 9008
    else:
        out = helper.create_variable(
            name=name, dtype=x.dtype, persistable=False)
S
sneaxiy 已提交
9009

S
sneaxiy 已提交
9010 9011 9012 9013 9014 9015 9016 9017 9018 9019 9020
    helper.append_op(
        type=op_type,
        inputs={'X': x,
                'Y': y},
        outputs={'Out': out},
        attrs={'axis': axis,
               'use_mkldnn': use_mkldnn})
    return helper.append_activation(out)


@templatedoc()
S
sneaxiy 已提交
9021
def scale(x, scale=1.0, bias=0.0, bias_after_scale=True, act=None, name=None):
S
sneaxiy 已提交
9022 9023 9024 9025 9026 9027 9028 9029
    """
    ${comment}

    Args:
        x(${x_type}): ${x_comment}
        scale(${scale_type}): ${scale_comment}
        bias(${bias_type}): ${bias_comment}
        bias_after_scale(${bias_after_scale_type}): ${bias_after_scale_comment}
S
sneaxiy 已提交
9030
        act(basestring|None): Activation applied to the output.
M
minqiyang 已提交
9031
        name(basestring|None): Name of the output.
S
sneaxiy 已提交
9032 9033 9034 9035 9036 9037

    Returns:
        out(${out_type}): ${out_comment}
    """

    helper = LayerHelper('scale', **locals())
S
sneaxiy 已提交
9038
    if name is None:
X
Xin Pan 已提交
9039
        out = helper.create_variable_for_type_inference(dtype=x.dtype)
S
sneaxiy 已提交
9040 9041 9042
    else:
        out = helper.create_variable(
            name=name, dtype=x.dtype, persistable=False)
S
sneaxiy 已提交
9043 9044 9045 9046 9047 9048 9049 9050 9051 9052

    helper.append_op(
        type='scale',
        inputs={'X': x},
        outputs={'Out': out},
        attrs={
            'scale': float(scale),
            'bias': float(bias),
            'bias_after_scale': bias_after_scale
        })
S
sneaxiy 已提交
9053
    return helper.append_activation(out)
S
sneaxiy 已提交
9054 9055


X
Xin Pan 已提交
9056
def elementwise_add(x, y, axis=-1, act=None, name=None):
S
sneaxiy 已提交
9057 9058 9059
    return _elementwise_op(LayerHelper('elementwise_add', **locals()))


X
Xin Pan 已提交
9060
def elementwise_div(x, y, axis=-1, act=None, name=None):
S
sneaxiy 已提交
9061 9062 9063
    return _elementwise_op(LayerHelper('elementwise_div', **locals()))


X
Xin Pan 已提交
9064
def elementwise_sub(x, y, axis=-1, act=None, name=None):
S
sneaxiy 已提交
9065 9066 9067
    return _elementwise_op(LayerHelper('elementwise_sub', **locals()))


X
Xin Pan 已提交
9068
def elementwise_mul(x, y, axis=-1, act=None, name=None):
S
sneaxiy 已提交
9069 9070 9071
    return _elementwise_op(LayerHelper('elementwise_mul', **locals()))


X
Xin Pan 已提交
9072
def elementwise_max(x, y, axis=-1, act=None, name=None):
S
sneaxiy 已提交
9073 9074 9075
    return _elementwise_op(LayerHelper('elementwise_max', **locals()))


X
Xin Pan 已提交
9076
def elementwise_min(x, y, axis=-1, act=None, name=None):
S
sneaxiy 已提交
9077 9078 9079
    return _elementwise_op(LayerHelper('elementwise_min', **locals()))


X
Xin Pan 已提交
9080
def elementwise_pow(x, y, axis=-1, act=None, name=None):
S
sneaxiy 已提交
9081 9082 9083 9084 9085 9086 9087 9088 9089 9090 9091
    return _elementwise_op(LayerHelper('elementwise_pow', **locals()))


for func in [
        elementwise_add, elementwise_div, elementwise_sub, elementwise_mul,
        elementwise_max, elementwise_min, elementwise_pow
]:
    op_proto = OpProtoHolder.instance().get_op_proto(func.__name__)
    func.__doc__ = _generate_doc_string_(
        op_proto,
        additional_args_lines=[
S
sneaxiy 已提交
9092 9093
            "act (basestring|None): Activation applied to the output.",
            "name (basestring|None): Name of the output."
S
sneaxiy 已提交
9094
        ])
M
minqiyang 已提交
9095 9096


9097
def _logical_op(op_name, x, y, out=None, name=None, binary_op=True):
M
minqiyang 已提交
9098 9099
    helper = LayerHelper(op_name, **locals())

M
minqiyang 已提交
9100 9101
    if binary_op:
        assert x.dtype == y.dtype
M
minqiyang 已提交
9102 9103 9104

    if out is None:
        if name is None:
X
Xin Pan 已提交
9105
            out = helper.create_variable_for_type_inference(dtype=x.dtype)
M
minqiyang 已提交
9106 9107 9108 9109 9110 9111 9112 9113 9114 9115 9116 9117 9118 9119 9120
        else:
            out = helper.create_variable(
                name=name, dtype=x.dtype, persistable=False)

    if binary_op:
        helper.append_op(
            type=op_name, inputs={"X": x,
                                  "Y": y}, outputs={"Out": out})
    else:
        helper.append_op(type=op_name, inputs={"X": x}, outputs={"Out": out})

    return out


@templatedoc()
9121
def logical_and(x, y, out=None, name=None):
M
minqiyang 已提交
9122 9123 9124 9125 9126 9127 9128 9129 9130 9131 9132
    """
    ${comment}

    Args:
        x(${x_type}): ${x_comment}
        y(${y_type}): ${y_comment}
        out(Tensor): Output tensor of logical operation.
        name(basestring|None): Name of the output.

    Returns:
        out(${out_type}): ${out_comment}
9133 9134 9135 9136 9137 9138 9139 9140 9141

    Examples:
        .. code-block:: python

            left = fluid.layers.data(
                name='left', shape=[1], dtype='int32')
            right = fluid.layers.data(
                name='right', shape=[1], dtype='int32')
            result = fluid.layers.logical_and(x=left, y=right)
M
minqiyang 已提交
9142 9143 9144 9145 9146 9147 9148
    """

    return _logical_op(
        op_name="logical_and", x=x, y=y, name=name, out=out, binary_op=True)


@templatedoc()
9149
def logical_or(x, y, out=None, name=None):
M
minqiyang 已提交
9150 9151 9152 9153 9154 9155 9156 9157 9158 9159 9160
    """
    ${comment}

    Args:
        x(${x_type}): ${x_comment}
        y(${y_type}): ${y_comment}
        out(Tensor): Output tensor of logical operation.
        name(basestring|None): Name of the output.

    Returns:
        out(${out_type}): ${out_comment}
9161 9162 9163 9164 9165 9166 9167 9168 9169

    Examples:
        .. code-block:: python

            left = fluid.layers.data(
                name='left', shape=[1], dtype='int32')
            right = fluid.layers.data(
                name='right', shape=[1], dtype='int32')
            result = fluid.layers.logical_or(x=left, y=right)
M
minqiyang 已提交
9170 9171 9172 9173 9174 9175 9176
    """

    return _logical_op(
        op_name="logical_or", x=x, y=y, name=name, out=out, binary_op=True)


@templatedoc()
9177
def logical_xor(x, y, out=None, name=None):
M
minqiyang 已提交
9178 9179 9180 9181 9182 9183 9184 9185 9186 9187 9188
    """
    ${comment}

    Args:
        x(${x_type}): ${x_comment}
        y(${y_type}): ${y_comment}
        out(Tensor): Output tensor of logical operation.
        name(basestring|None): Name of the output.

    Returns:
        out(${out_type}): ${out_comment}
9189 9190 9191 9192 9193 9194 9195 9196 9197

    Examples:
        .. code-block:: python

            left = fluid.layers.data(
                name='left', shape=[1], dtype='int32')
            right = fluid.layers.data(
                name='right', shape=[1], dtype='int32')
            result = fluid.layers.logical_xor(x=left, y=right)
M
minqiyang 已提交
9198 9199 9200 9201 9202 9203 9204
    """

    return _logical_op(
        op_name="logical_xor", x=x, y=y, name=name, out=out, binary_op=True)


@templatedoc()
9205
def logical_not(x, out=None, name=None):
M
minqiyang 已提交
9206 9207 9208 9209 9210 9211 9212 9213 9214 9215
    """
    ${comment}

    Args:
        x(${x_type}): ${x_comment}
        out(Tensor): Output tensor of logical operation.
        name(basestring|None): Name of the output.

    Returns:
        out(${out_type}): ${out_comment}
9216 9217 9218 9219 9220 9221 9222

    Examples:
        .. code-block:: python

            left = fluid.layers.data(
                name='left', shape=[1], dtype='int32')
            result = fluid.layers.logical_not(x=left)
M
minqiyang 已提交
9223 9224 9225 9226
    """

    return _logical_op(
        op_name="logical_not", x=x, y=None, name=name, out=out, binary_op=False)
9227 9228 9229 9230 9231 9232 9233 9234 9235 9236 9237 9238 9239 9240 9241


@templatedoc()
def clip(x, min, max, name=None):
    """
    ${comment}

    Args:
        x(${x_type}): ${x_comment}
        min(${min_type}): ${min_comment}
        max(${max_type}): ${max_comment}
        name(basestring|None): Name of the output.

    Returns:
        out(${out_type}): ${out_comment}
9242 9243 9244 9245 9246 9247 9248

    Examples:
        .. code-block:: python

            input = fluid.layers.data(
                name='data', shape=[1], dtype='float32')
            reward = fluid.layers.clip(x=input, min=-1.0, max=1.0)
9249 9250 9251 9252 9253
    """

    helper = LayerHelper("clip", **locals())

    if name is None:
S
sneaxiy 已提交
9254 9255 9256 9257
        name = unique_name.generate(".".join([helper.name, 'tmp']))

    out = helper.create_variable(
        type=x.type, name=name, dtype=x.dtype, persistable=False)
9258 9259 9260 9261 9262 9263 9264 9265 9266 9267 9268 9269 9270 9271 9272 9273 9274 9275 9276 9277 9278 9279 9280

    helper.append_op(
        type="clip",
        inputs={"X": x},
        attrs={"min": min,
               "max": max},
        outputs={"Out": out})

    return out


@templatedoc()
def clip_by_norm(x, max_norm, name=None):
    """
    ${comment}

    Args:
        x(${x_type}): ${x_comment}
        max_norm(${max_norm_type}): ${max_norm_comment}
        name(basestring|None): Name of the output.

    Returns:
        out(${out_type}): ${out_comment}
9281 9282 9283 9284 9285 9286 9287

    Examples:
        .. code-block:: python

            input = fluid.layers.data(
                name='data', shape=[1], dtype='float32')
            reward = fluid.layers.clip_by_norm(x=input, max_norm=1.0)
9288 9289 9290 9291 9292
    """

    helper = LayerHelper("clip_by_norm", **locals())

    if name is None:
S
sneaxiy 已提交
9293 9294 9295 9296
        name = unique_name.generate(".".join([helper.name, 'tmp']))

    out = helper.create_variable(
        type=x.type, name=name, dtype=x.dtype, persistable=False)
9297 9298 9299 9300 9301 9302 9303 9304

    helper.append_op(
        type="clip_by_norm",
        inputs={"X": x},
        attrs={"max_norm": max_norm},
        outputs={"Out": out})

    return out
X
Xin Pan 已提交
9305 9306 9307 9308 9309 9310 9311 9312 9313 9314 9315 9316 9317 9318 9319 9320 9321 9322


@templatedoc()
def mean(x, name=None):
    """
    ${comment}

    Args:
        x(${x_type}): ${x_comment}
        name(basestring|None): Name of the output.

    Returns:
        out(${out_type}): ${out_comment}
    """

    helper = LayerHelper("mean", **locals())

    if name is None:
X
Xin Pan 已提交
9323
        out = helper.create_variable_for_type_inference(dtype=x.dtype)
X
Xin Pan 已提交
9324 9325 9326 9327 9328 9329 9330 9331 9332 9333
    else:
        out = helper.create_variable(
            name=name, dtype=x.dtype, persistable=False)

    helper.append_op(
        type="mean", inputs={"X": x}, attrs={}, outputs={"Out": out})

    return out


C
chengduo 已提交
9334 9335 9336 9337 9338 9339 9340 9341 9342 9343 9344 9345 9346 9347 9348 9349 9350 9351 9352 9353 9354 9355 9356
@templatedoc()
def merge_selected_rows(x, name=None):
    """
    ${comment}

    Args:
        x(${x_type}): ${x_comment}
        name(basestring|None): Name of the output.

    Returns:
        out(${out_type}): ${out_comment}
    """

    helper = LayerHelper("merge_selected_rows", **locals())
    out = helper.create_variable_for_type_inference(dtype=x.dtype)
    helper.append_op(
        type="merge_selected_rows",
        inputs={"X": x},
        attrs={},
        outputs={"Out": out})
    return out


X
Xin Pan 已提交
9357 9358 9359 9360 9361 9362 9363 9364 9365 9366 9367 9368 9369 9370 9371 9372 9373 9374 9375
@templatedoc()
def mul(x, y, x_num_col_dims=1, y_num_col_dims=1, name=None):
    """
    ${comment}

    Args:
        x(${x_type}): ${x_comment}
        y(${y_type}): ${y_comment}
        x_num_col_dims(${x_num_col_dims_type}): ${x_num_col_dims_comment}
        y_num_col_dims(${y_num_col_dims_type}): ${y_num_col_dims_comment}
        name(basestring|None): Name of the output.

    Returns:
        out(${out_type}): ${out_comment}
    """

    helper = LayerHelper("mul", **locals())

    if name is None:
X
Xin Pan 已提交
9376
        out = helper.create_variable_for_type_inference(dtype=x.dtype)
X
Xin Pan 已提交
9377 9378 9379 9380 9381 9382 9383 9384 9385
    else:
        out = helper.create_variable(
            name=name, dtype=x.dtype, persistable=False)

    helper.append_op(
        type="mul",
        inputs={"X": x,
                "Y": y},
        attrs={
X
fix  
Xin Pan 已提交
9386 9387
            "x_num_col_dims": x_num_col_dims,
            "y_num_col_dims": y_num_col_dims
X
Xin Pan 已提交
9388 9389 9390 9391 9392 9393
        },
        outputs={"Out": out})
    return out


@templatedoc()
J
jerrywgz 已提交
9394 9395 9396
def sigmoid_cross_entropy_with_logits(x,
                                      label,
                                      ignore_index=kIgnoreIndex,
9397 9398
                                      name=None,
                                      normalize=False):
X
Xin Pan 已提交
9399 9400 9401 9402 9403 9404
    """
    ${comment}

    Args:
        x(${x_type}): ${x_comment}
        label(${label_type}): ${label_comment}
9405
        ignore_index(&{ignore_index}): ${ignore_index_comment}
X
Xin Pan 已提交
9406
        name(basestring|None): Name of the output.
9407 9408
        normalize(bool): If true, divide the output by the number of
            targets != ignore_index.
X
Xin Pan 已提交
9409 9410 9411

    Returns:
        out(${out_type}): ${out_comment}
9412 9413 9414 9415 9416 9417 9418 9419 9420 9421 9422 9423 9424 9425

    Examples:
        .. code-block:: python

            input = fluid.layers.data(
                name='data', shape=[10], dtype='float32')
            label = fluid.layers.data(
                name='data', shape=[10], dtype='float32')
            loss = fluid.layers.sigmoid_cross_entropy_with_logits(
                x=input,
                label=label,
                ignore_index=-1,
                normalize=True) # or False
            # loss = fluid.layers.reduce_sum(loss) # summation of loss
X
Xin Pan 已提交
9426 9427 9428 9429 9430
    """

    helper = LayerHelper("sigmoid_cross_entropy_with_logits", **locals())

    if name is None:
X
Xin Pan 已提交
9431
        out = helper.create_variable_for_type_inference(dtype=x.dtype)
X
Xin Pan 已提交
9432 9433 9434 9435 9436 9437 9438 9439
    else:
        out = helper.create_variable(
            name=name, dtype=x.dtype, persistable=False)

    helper.append_op(
        type="sigmoid_cross_entropy_with_logits",
        inputs={"X": x,
                "Label": label},
9440 9441
        attrs={"ignore_index": ignore_index,
               'normalize': normalize},
X
Xin Pan 已提交
9442 9443 9444 9445 9446 9447 9448 9449 9450 9451 9452 9453 9454 9455 9456 9457 9458 9459 9460 9461
        outputs={"Out": out})
    return out


@templatedoc()
def maxout(x, groups, name=None):
    """
    ${comment}

    Args:
        x(${x_type}): ${x_comment}
        groups(${groups_type}): ${groups_comment}
        name(basestring|None): Name of the output.

    Returns:
        out(${out_type}): ${out_comment}
    """
    helper = LayerHelper("maxout", **locals())

    if name is None:
X
Xin Pan 已提交
9462
        out = helper.create_variable_for_type_inference(dtype=x.dtype)
X
Xin Pan 已提交
9463 9464 9465 9466 9467 9468 9469 9470 9471 9472
    else:
        out = helper.create_variable(
            name=name, dtype=x.dtype, persistable=False)

    helper.append_op(
        type="maxout",
        inputs={"X": x},
        attrs={"groups": groups},
        outputs={"Out": out})
    return out
9473 9474


J
JiabinYang 已提交
9475
def space_to_depth(x, blocksize, name=None):
J
JiabinYang 已提交
9476
    """
J
JiabinYang 已提交
9477
    Gives a blocksize to space_to_depth the input LoDtensor with Layout: [batch, channel, height, width]
9478 9479 9480

    This op rearranges blocks of spatial data, into depth. More specifically, this op outputs a copy of the
    input LoDtensor where values from the height and width dimensions are moved to the channel dimension.
J
JiabinYang 已提交
9481
    The attr blocksize indicates the input block size.
9482 9483

    space_to_depth will reorgnize the elements of input with shape[batch, channel, height, width] according
J
JiabinYang 已提交
9484
    to blocksize to construct output with shape [batch, channel * blocksize * blocksize, height/blocksize, width/blocksize]:
9485 9486

    space_to_depth is used to This operation is useful for resizing the activations between convolutions
J
JiabinYang 已提交
9487
    (but keeping all data)
J
JiabinYang 已提交
9488

J
JiabinYang 已提交
9489
    - Non-overlapping blocks of size block_size x block size are rearranged into depth at each location.
9490
    - The depth of the output tensor is block_size * block_size * input channel
J
JiabinYang 已提交
9491 9492 9493 9494 9495
    - The Y, X coordinates within each block of the input become the high order component of the output channel index
    - channel should be divisible by square of blocksize
    - height, width should be divsible by blocksize


J
JiabinYang 已提交
9496
    Args:
J
JiabinYang 已提交
9497
        x(variable): The input LoDtensor.
J
JiabinYang 已提交
9498
        blocksize(variable): The blocksize to select the element on each feature map should be > 2
J
JiabinYang 已提交
9499 9500

    Returns:
J
JiabinYang 已提交
9501
        Variable: The output LoDtensor.
J
JiabinYang 已提交
9502 9503

    Raises:
J
JiabinYang 已提交
9504
        TypeError: blocksize type must be a long.
J
JiabinYang 已提交
9505 9506 9507 9508 9509 9510

    Examples:
        .. code-block:: python

            data = fluid.layers.data(
                name='data', shape=[1, 4, 2, 2], dtype='float32')
J
JiabinYang 已提交
9511
            space_to_depthed = fluid.layers.space_to_depth(
J
JiabinYang 已提交
9512
                x=data, blocksize=2)
J
JiabinYang 已提交
9513 9514
    """

J
JiabinYang 已提交
9515
    helper = LayerHelper("space_to_depth", **locals())
J
JiabinYang 已提交
9516

J
JiabinYang 已提交
9517 9518
    if not (isinstance(blocksize, int)):
        raise ValueError("blocksize must be a python Int")
J
JiabinYang 已提交
9519 9520

    if name is None:
J
JiabinYang 已提交
9521 9522
        out = helper.create_variable_for_type_inference(
            dtype=x.dtype)  #fix create
J
JiabinYang 已提交
9523 9524 9525 9526 9527
    else:
        out = helper.create_variable(
            name=name, dtype=x.dtype, persistable=False)

    helper.append_op(
J
JiabinYang 已提交
9528
        type="space_to_depth",
J
JiabinYang 已提交
9529
        inputs={"X": x},
J
JiabinYang 已提交
9530
        attrs={"blocksize": blocksize},
J
JiabinYang 已提交
9531
        outputs={"Out": out})
J
JiabinYang 已提交
9532 9533
    return out

J
JiabinYang 已提交
9534

S
sneaxiy 已提交
9535 9536
@templatedoc()
def sequence_reverse(x, name=None):
9537
    """
S
sneaxiy 已提交
9538 9539 9540 9541 9542 9543 9544 9545 9546 9547 9548
    ${comment}

    Args:
        x(${x_type}): ${x_comment}
        name(basestring|None): Name of the output.

    Returns:
        out(${y_type}): ${y_comment}
    """
    helper = LayerHelper("sequence_reverse", **locals())
    if name is None:
S
sneaxiy 已提交
9549
        out = helper.create_variable_for_type_inference(dtype=x.dtype)
S
sneaxiy 已提交
9550 9551 9552 9553 9554 9555 9556 9557 9558 9559
    else:
        out = helper.create_variable(
            name=name, dtype=x.dtype, persistable=False)

    helper.append_op(
        type="sequence_reverse",
        inputs={"X": x},
        outputs={"Y": out},
        attrs=dict())
    return out
S
sneaxiy 已提交
9560 9561


9562 9563 9564 9565 9566 9567
def affine_channel(x, scale=None, bias=None, data_layout='NCHW', name=None):
    """
    Applies a separate affine transformation to each channel of the input.
    Useful for replacing spatial batch norm with its equivalent fixed
    transformation. The input also can be 2D tensor and applies a affine
    transformation in second dimension.
9568

9569 9570 9571 9572 9573 9574 9575 9576 9577 9578 9579 9580 9581 9582 9583 9584 9585 9586 9587
    Args:
        x (Variable): Feature map input can be a 4D tensor with order NCHW
            or NHWC. It also can be a 2D tensor and the affine transformation
            is applied in the second dimension.
        scale (Variable): 1D input of shape (C), the c-th element is the scale
            factor of the affine transformation for the c-th channel of
            the input.
        bias (Variable): 1D input of shape (C), the c-th element is the bias
            of the affine transformation for the c-th channel of the input.
        data_layout (string, default NCHW): NCHW or NHWC. If input is 2D
            tensor, you can ignore data_layout.
        name (str, default None): The name of this layer.

    Returns:
        out (Variable): A tensor of the same shape and data layout with x.
    """
    helper = LayerHelper("affine_channel", **locals())

    if name is None:
X
Xin Pan 已提交
9588
        out = helper.create_variable_for_type_inference(dtype=x.dtype)
9589 9590 9591 9592 9593 9594 9595 9596 9597 9598 9599 9600
    else:
        out = helper.create_variable(
            name=name, dtype=x.dtype, persistable=False)

    helper.append_op(
        type="affine_channel",
        inputs={"X": x,
                'Scale': scale,
                'Bias': bias},
        attrs={"data_layout": data_layout},
        outputs={"Out": out})
    return out
9601 9602


B
barrierye 已提交
9603
def similarity_focus(input, axis, indexes, name=None):
9604
    """
B
barrierye 已提交
9605
    SimilarityFocus Operator
B
barrierye 已提交
9606 9607

    Generate a similarity focus mask with the same shape of input using the following method:
M
minqiyang 已提交
9608

9609 9610 9611
    1. Extract the 3-D tensor(here the first dimension is BatchSize) corresponding
       to the axis according to the indexes. For example, if axis=1 and indexes=[a],
       it will get the matrix T=X[:, a, :, :]. In this case, if the shape of input X
B
barrierye 已提交
9612
       is (BatchSize, A, B, C), the shape of tensor T is (BatchSize, B, C).
9613 9614 9615 9616 9617 9618 9619
    2. For each index, find the largest numbers in the tensor T, so that the same
       row and same column has at most one number(what it means is that if the
       largest number has been found in the i-th row and the j-th column, then
       the numbers in the i-th row or j-th column will be skipped. And then the
       next largest number will be selected from the remaining numbers. Obviously
       there will be min(B, C) numbers), and mark the corresponding position of the
       3-D similarity focus mask as 1, otherwise as 0. Do elementwise-or for
B
barrierye 已提交
9620
       each index.
B
barrierye 已提交
9621 9622 9623 9624
    3. Broadcast the 3-D similarity focus mask to the same shape of input X.

    Refer to `Similarity Focus Layer <http://www.aclweb.org/anthology/N16-1108>`_

B
barrierye 已提交
9625 9626 9627 9628 9629 9630 9631 9632 9633 9634 9635 9636 9637 9638 9639 9640 9641 9642 9643 9644 9645 9646 9647 9648 9649 9650 9651 9652 9653 9654 9655 9656 9657 9658 9659 9660 9661 9662 9663 9664 9665 9666 9667 9668 9669 9670 9671 9672 9673
    .. code-block:: text

        * Example :

            Given a 4-D tensor x with the shape (BatchSize, C, A, B), where C is
            the number of channels and the shape of feature map is (A, B):
                x.shape = (2, 3, 2, 2)
                x.data = [[[[0.8, 0.1],
                            [0.4, 0.5]],

                           [[0.9, 0.7],
                            [0.9, 0.9]],

                           [[0.8, 0.9],
                            [0.1, 0.2]]],


                          [[[0.2, 0.5],
                            [0.3, 0.4]],

                           [[0.9, 0.7],
                            [0.8, 0.4]],

                           [[0.0, 0.2],
                            [0.4, 0.7]]]]

            Given axis: 1 (the axis of the channel)
            Given indexes: [0]

            then we get a 4-D tensor out with the same shape of input x:
                out.shape = (2, 3, 2, 2)
                out.data = [[[[1.0, 0.0],
                              [0.0, 1.0]],

                             [[1.0, 0.0],
                              [0.0, 1.0]],

                             [[1.0, 0.0],
                              [0.0, 1.0]]],

                            [[[0.0, 1.0],
                              [1.0, 0.0]],

                             [[0.0, 1.0],
                              [1.0, 0.0]],

                             [[0.0, 1.0],
                              [1.0, 0.0]]]]

B
barrierye 已提交
9674
    Args:
9675
        input(Variable): The input tensor variable(default float). It should
B
barrierye 已提交
9676
            be a 4-D tensor with shape [BatchSize, A, B, C].
B
barrierye 已提交
9677
        axis(int): Indicating the dimension to be selected. It can only be
B
barrierye 已提交
9678
            1, 2 or 3.
B
barrierye 已提交
9679
        indexes(list): Indicating the indexes of the selected dimension.
B
barrierye 已提交
9680 9681

    Returns:
H
haowang101779990 已提交
9682 9683
        Variable: A tensor variable with the same shape and same type \
                  as the input.
9684

B
barrierye 已提交
9685 9686
    Examples:
        .. code-block:: python
H
haowang101779990 已提交
9687

B
barrierye 已提交
9688
            data = fluid.layers.data(
B
barrierye 已提交
9689 9690
              name='data', shape=[2, 3, 2, 2], dtype='float32')
            x = fluid.layers.layer_norm(input=data, axis=1, indexes=[0])
H
haowang101779990 已提交
9691

B
barrierye 已提交
9692 9693 9694 9695 9696 9697 9698 9699 9700 9701 9702 9703
    """
    helper = LayerHelper('similarity_focus', **locals())
    # check attrs
    if isinstance(axis, int) is False:
        raise TypeError("axis must be int type.")
    if isinstance(indexes, list) is False:
        raise TypeError("indexes must be list type.")
    if axis != 1 and axis != 2 and axis != 3:
        raise ValueError("axis must be 1, 2 or 3.")
    if len(indexes) == 0:
        raise ValueError("indexes can not be empty.")

B
barrierye 已提交
9704 9705 9706 9707 9708
    if name is None:
        out = helper.create_variable_for_type_inference(dtype=input.dtype)
    else:
        out = helper.create_variable(
            name=name, dtype=input.dtype, persistable=False)
B
barrierye 已提交
9709 9710 9711 9712 9713 9714 9715
    helper.append_op(
        type='similarity_focus',
        inputs={'X': input},
        outputs={'Out': out},
        attrs={"axis": axis,
               "indexes": indexes})
    return out
B
barrierye 已提交
9716 9717


M
minqiyang 已提交
9718 9719
def hash(input, hash_size, num_hash=1, name=None):
    """
M
minqiyang 已提交
9720 9721
    Hash the input to an integer whose value is less than the given hash size.

M
minqiyang 已提交
9722 9723
    The hash algorithm we used was xxHash - Extremely fast hash algorithm
    (https://github.com/Cyan4973/xxHash/tree/v0.6.5)
M
minqiyang 已提交
9724 9725 9726 9727 9728 9729 9730 9731 9732 9733 9734 9735 9736 9737 9738 9739 9740 9741 9742 9743 9744 9745 9746 9747 9748 9749 9750 9751 9752 9753 9754 9755 9756 9757 9758 9759 9760 9761

    A simple example as below:

    .. code-block:: text

        Given:

        # shape [2, 2]
        input.data = [
            [[1], [2]],
            [[3], [4]],
        ]

        input.lod = [[0, 2]]

        hash_size = 10000

        num_hash = 4

        Then:

        Hash op will take all number in input's 2nd dimension as hash algorithm's
        input for each time. Each input will be hashed for 4 times, and get an
        array whose length is 4. Each value in the array ranges from 0 to 9999.

        # shape [2, 4]
        output.data = [
            [[9662], [9217], [1129], [8487]],
            [[8310], [1327], [1654], [4567]],
        ]

        output.lod = [[0, 2]]

    Args:
        input (Variable): The input variable which is a one-hot word. The
            dimensions of the input variable must be 2.
        hash_size (int): The space size for hash algorithm. The output value
            will keep in the range:math:`[0, hash_size - 1]`.
M
minqiyang 已提交
9762
        num_hash (int): The times of hash, default 1.
M
minqiyang 已提交
9763
        name (str, default None): The name of this layer.
M
minqiyang 已提交
9764 9765 9766 9767 9768 9769

    Returns:
       Variable: The hash result variable which is a LoDTensor.

    Examples:
       .. code-block:: python
H
haowang101779990 已提交
9770

M
minqiyang 已提交
9771 9772 9773
           word_dict = paddle.dataset.imdb.word_dict()
           x = fluid.layers.data(shape[1], dtype='int32', lod_level=1)
           out = fluid.layers.hash(input=x, num_hash=4, hash_size=1000)
M
minqiyang 已提交
9774 9775
    """
    helper = LayerHelper('hash', **locals())
M
minqiyang 已提交
9776 9777
    out = helper.create_variable_for_type_inference(
        helper.input_dtype(), stop_gradient=True)
M
minqiyang 已提交
9778 9779 9780 9781 9782 9783 9784
    helper.append_op(
        type='hash',
        inputs={'X': input},
        outputs={'Out': out},
        attrs={'num_hash': num_hash,
               'mod_by': hash_size})
    return out
G
gmcather 已提交
9785 9786


D
dengkaipeng 已提交
9787
@templatedoc()
9788 9789
def grid_sampler(x, grid, name=None):
    """
9790
    This operation samples input X by using bilinear interpolation based on
H
haowang101779990 已提交
9791
    flow field grid, which is usually gennerated by :code:`affine_grid` . The grid of
9792 9793 9794 9795
    shape [N, H, W, 2] is the concatenation of (grid_x, grid_y) coordinates
    with shape [N, H, W] each, where grid_x is indexing the 4th dimension
    (in width dimension) of input data x and grid_y is indexng the 3rd
    dimention (in height dimension), finally results is the bilinear
9796
    interpolation value of 4 nearest corner points.
9797

H
haowang101779990 已提交
9798
    .. code-block:: text
9799

H
haowang101779990 已提交
9800 9801
        Step 1:
        Get (x, y) grid coordinates and scale to [0, H-1/W-1].
9802

H
haowang101779990 已提交
9803 9804
        grid_x = 0.5 * (grid[:, :, :, 0] + 1) * (W - 1)
        grid_y = 0.5 * (grid[:, :, :, 1] + 1) * (H - 1)
9805

H
haowang101779990 已提交
9806 9807 9808
        Step 2:
        Indices input data X with grid (x, y) in each [H, W] area, and bilinear
        interpolate point value by 4 nearest points.
9809

H
haowang101779990 已提交
9810 9811 9812 9813 9814 9815 9816 9817 9818
          wn ------- y_n ------- en
          |           |           |
          |          d_n          |
          |           |           |
         x_w --d_w-- grid--d_e-- x_e
          |           |           |
          |          d_s          |
          |           |           |
          ws ------- y_s ------- wn
9819

H
haowang101779990 已提交
9820 9821 9822 9823
        x_w = floor(x)              // west side x coord
        x_e = x_w + 1               // east side x coord
        y_n = floor(y)              // north side y coord
        y_s = y_s + 1               // south side y coord
9824

H
haowang101779990 已提交
9825 9826 9827 9828
        d_w = grid_x - x_w          // distance to west side
        d_e = x_e - grid_x          // distance to east side
        d_n = grid_y - y_n          // distance to north side
        d_s = y_s - grid_y          // distance to south side
9829

H
haowang101779990 已提交
9830 9831 9832 9833
        wn = X[:, :, y_n, x_w]      // north-west point value
        en = X[:, :, y_n, x_e]      // north-east point value
        ws = X[:, :, y_s, x_w]      // south-east point value
        es = X[:, :, y_s, x_w]      // north-east point value
9834

H
haowang101779990 已提交
9835 9836
        output = wn * d_e * d_s + en * d_w * d_s
               + ws * d_e * d_n + es * d_w * d_n
D
dengkaipeng 已提交
9837 9838

    Args:
9839 9840 9841
        x(Variable): Input data of shape [N, C, H, W].
        grid(Variable): Input grid tensor of shape [N, H, W, 2].
        name (str, default None): The name of this layer.
D
dengkaipeng 已提交
9842 9843

    Returns:
H
haowang101779990 已提交
9844
        Variable: Output of shape [N, C, H, W] data samples input X
9845 9846
        using bilnear interpolation based on input grid.

H
haowang101779990 已提交
9847 9848 9849 9850 9851 9852 9853 9854
    Examples:

        .. code-block:: python

            x = fluid.layers.data(name='x', shape=[3, 10, 32, 32], dtype='float32')
            theta = fluid.layers.data(name='theta', shape=[3, 2, 3], dtype='float32')
            grid = fluid.layers.affine_grid(input=theta, size=[3, 10, 32, 32]})
            out = fluid.layers.grid_sampler(x=x, grid=grid)
9855

D
dengkaipeng 已提交
9856 9857 9858 9859 9860 9861 9862 9863 9864
    """
    helper = LayerHelper("grid_sampler", **locals())

    if not isinstance(x, Variable):
        return ValueError("The x should be a Variable")

    if not isinstance(grid, Variable):
        return ValueError("The grid should be a Variable")

9865
    out = helper.create_variable_for_type_inference(x.dtype)
D
dengkaipeng 已提交
9866 9867
    ipts = {'X': x, 'Grid': grid}

9868
    helper.append_op(type='grid_sampler', inputs=ipts, outputs={'Output': out})
9869 9870 9871
    return out


G
gmcather 已提交
9872 9873 9874 9875 9876 9877 9878 9879 9880 9881 9882 9883 9884 9885 9886 9887 9888 9889 9890 9891 9892 9893 9894 9895 9896 9897 9898 9899 9900 9901 9902 9903 9904 9905 9906 9907 9908 9909 9910 9911 9912 9913 9914 9915 9916 9917 9918
def log_loss(input, label, epsilon=1e-4, name=None):
    """
    **Negative Log Loss Layer**

    This layer accepts input predictions and target label and returns the
    negative log loss.

    .. math::

        Out = -label * \\log{(input + \\epsilon)}
              - (1 - label) * \\log{(1 - input + \\epsilon)}

    Args:
        input (Variable|list):  a 2-D tensor with shape [N x 1], where N is the
                                batch size. This input is a probability computed
                                by the previous operator.
        label (Variable|list):  the ground truth which is a 2-D tensor with
                                shape [N x 1], where N is the batch size.
        epsilon (float): epsilon
        name (string): the name of log_loss

    Returns:
        Variable: A 2-D tensor with shape [N x 1], the negative log loss.

    Examples:
        .. code-block:: python

          prob = fluid.layers.sigmoid(net)
          cost = fluid.layers.log_loss(input=prob, label=label)
    """
    helper = LayerHelper('log_loss', **locals())

    if name is None:
        loss = helper.create_variable_for_type_inference(dtype=input.dtype)
    else:
        loss = helper.create_variable(
            name=name, dtype=input.dtype, persistable=False)

    helper.append_op(
        type='log_loss',
        inputs={'Predicted': [input],
                'Labels': [label]},
        outputs={'Loss': [loss]},
        attrs={'epsilon': epsilon})
    return loss


H
heqiaozhi 已提交
9919 9920 9921 9922 9923 9924 9925 9926 9927 9928 9929 9930 9931 9932 9933 9934 9935 9936 9937
def teacher_student_sigmoid_loss(input,
                                 label,
                                 soft_max_up_bound=15.0,
                                 soft_max_lower_bound=-15.0):
    """
    **Teacher Student Log Loss Layer**

    This layer accepts input predictions and target label and returns the
    teacher_student loss.

    .. math::
        loss = max(x, 0) - x * z + log(1 + exp(-abs(x))) + max(x, 0) - x * z' + log(1 + exp(-abs(x)))

    Args:
        input (Variable|list):  a 2-D tensor with shape [N x 1], where N is the
                                batch size. This input is a probability computed
                                by the previous operator.
        label (Variable|list):  the ground truth which is a 2-D tensor with
                                shape [N x 1], where N is the batch size.
M
minqiyang 已提交
9938
        soft_max_up_bound  (float):  if input > soft_max_up_bound, will be bound
H
heqiaozhi 已提交
9939 9940 9941 9942 9943 9944 9945
        soft_max_lower_bound (float): if input < soft_max_lower_bound, will be bound

    Returns:
        Variable: A 2-D tensor with shape [N x 1], the teacher_student_sigmoid_loss.

    Examples:
        .. code-block:: python
H
heqiaozhi 已提交
9946

H
heqiaozhi 已提交
9947 9948 9949 9950 9951 9952 9953 9954 9955 9956 9957 9958 9959 9960
          cost = fluid.layers.teacher_student_sigmoid_loss(input=similarity, label=label)
    """
    helper = LayerHelper('teacher_student_sigmoid_loss', **locals())
    out = helper.create_variable(dtype=input.dtype)
    helper.append_op(
        type='teacher_student_sigmoid_loss',
        inputs={'X': [input],
                'Label': [label]},
        outputs={'Y': [out]},
        attrs={"soft_max_lower_bound": float(soft_max_lower_bound), \
                "soft_max_up_bound": float(soft_max_up_bound)})
    return out


G
gmcather 已提交
9961 9962 9963 9964
def add_position_encoding(input, alpha, beta, name=None):
    """
    **Add Position Encoding Layer**

H
haowang101779990 已提交
9965
    This layer accepts an input 3D-Tensor of shape [N x M x P], and returns an
G
gmcather 已提交
9966 9967
    output Tensor of shape [N x M x P] with positional encoding value.

H
haowang101779990 已提交
9968
    Refer to `Attention Is All You Need <http://arxiv.org/pdf/1706.03762.pdf>`_ .
G
gmcather 已提交
9969 9970

    .. math::
H
haowang101779990 已提交
9971 9972 9973
        PE(pos, 2i) &= \\sin{(pos / 10000^{2i / P})}   \\\\
        PE(pos, 2i + 1) &= \\cos{(pos / 10000^{2i / P})}  \\\\
        Out(:, pos, i) &= \\alpha * input(:, pos, i) + \\beta * PE(pos, i)
G
gmcather 已提交
9974 9975

    Where:
H
haowang101779990 已提交
9976 9977
      - :math:`PE(pos, 2i)` : the increment for the number at even position
      - :math:`PE(pos, 2i + 1)` : the increment for the number at odd position
G
gmcather 已提交
9978 9979 9980 9981 9982 9983 9984 9985 9986 9987 9988 9989 9990 9991

    Args:
        input (Variable): 3-D input tensor with shape [N x M x P]
        alpha (float): multiple of Input Tensor
        beta (float): multiple of Positional Encoding Tensor
        name (string): the name of position encoding layer

    Returns:
        Variable: A 3-D Tensor of shape [N x M x P] with positional encoding.

    Examples:
        .. code-block:: python

          position_tensor = fluid.layers.add_position_encoding(input=tensor)
H
haowang101779990 已提交
9992

G
gmcather 已提交
9993 9994 9995 9996 9997 9998 9999 10000 10001 10002 10003 10004 10005 10006 10007 10008
    """
    helper = LayerHelper('add_position_encoding', **locals())
    dtype = helper.input_dtype()

    if name is None:
        out = helper.create_variable_for_type_inference(dtype=dtype)
    else:
        out = helper.create_variable(name=name, dtype=dtype, persistable=False)

    helper.append_op(
        type="add_position_encoding",
        inputs={"X": input},
        outputs={"Out": out},
        attrs={"alpha": alpha,
               "beta": beta})
    return out
Q
Qiao Longfei 已提交
10009 10010 10011 10012 10013 10014 10015 10016 10017 10018


def bilinear_tensor_product(x,
                            y,
                            size,
                            act=None,
                            name=None,
                            param_attr=None,
                            bias_attr=None):
    """
Q
Qiao Longfei 已提交
10019
    **Add Bilinear Tensor Product Layer**
Q
Qiao Longfei 已提交
10020

Q
Qiao Longfei 已提交
10021
    This layer performs bilinear tensor product on two inputs.
Q
Qiao Longfei 已提交
10022 10023 10024
    For example:

    .. math::
H
haowang101779990 已提交
10025
       out_{i} = x * W_{i} * {y^\mathrm{T}}, i=0,1,...,size-1
Q
Qiao Longfei 已提交
10026

Q
Qiao Longfei 已提交
10027
    In this formula:
10028 10029
      - :math:`x`: the first input contains M elements, shape is [batch_size, M].
      - :math:`y`: the second input contains N elements, shape is [batch_size, N].
Q
Qiao Longfei 已提交
10030
      - :math:`W_{i}`: the i-th learned weight, shape is [M, N]
H
haowang101779990 已提交
10031
      - :math:`out_{i}`: the i-th element of out, shape is [batch_size, size].
Q
Qiao Longfei 已提交
10032 10033 10034
      - :math:`y^\mathrm{T}`: the transpose of :math:`y_{2}`.

    Args:
10035 10036
        x (Variable): 2-D input tensor with shape [batch_size, M]
        y (Variable): 2-D input tensor with shape [batch_size, N]
Q
Qiao Longfei 已提交
10037 10038 10039
        size (int): The dimension of this layer.
        act (str, default None): Activation to be applied to the output of this layer.
        name (str, default None): The name of this layer.
Q
Qiao Longfei 已提交
10040
        param_attr (ParamAttr, default None): The parameter attribute for the learnable w.
Q
Qiao Longfei 已提交
10041
            parameters/weights of this layer.
Q
Qiao Longfei 已提交
10042
        bias_attr (ParamAttr, default None): The parameter attribute for the bias
Q
Qiao Longfei 已提交
10043 10044 10045 10046
            of this layer. If it is set to False, no bias will be added to the output units.
            If it is set to None, the bias is initialized zero. Default: None.

    Returns:
Q
Qiao Longfei 已提交
10047
        Variable: A 2-D Tensor of shape [batch_size, size].
Q
Qiao Longfei 已提交
10048 10049 10050 10051

    Examples:
        .. code-block:: python

Q
Qiao Longfei 已提交
10052
          tensor = bilinear_tensor_product(x=layer1, y=layer2, size=1000)
Q
Qiao Longfei 已提交
10053 10054
    """
    helper = LayerHelper('bilinear_tensor_product', **locals())
Q
Qiao Longfei 已提交
10055
    dtype = helper.input_dtype('x')
Q
Qiao Longfei 已提交
10056 10057 10058 10059

    param_shape = [size, x.shape[1], y.shape[1]]

    w = helper.create_parameter(
Q
Qiao Longfei 已提交
10060
        attr=helper.param_attr, shape=param_shape, dtype=dtype, is_bias=False)
Q
Qiao Longfei 已提交
10061 10062 10063 10064 10065 10066 10067 10068 10069 10070 10071 10072 10073 10074 10075 10076 10077

    if name is None:
        out = helper.create_variable_for_type_inference(dtype=dtype)
    else:
        out = helper.create_variable(name=name, dtype=dtype, persistable=False)

    inputs = {"X": x, "Y": y, "Weight": w}
    if helper.bias_attr:
        bias_size = [1, size]
        bias = helper.create_parameter(
            attr=helper.bias_attr, shape=bias_size, dtype=dtype, is_bias=True)
        inputs["Bias"] = bias
    helper.append_op(
        type="bilinear_tensor_product", inputs=inputs, outputs={"Out": out})

    # add activation
    return helper.append_activation(out)
C
chengduo 已提交
10078 10079 10080 10081 10082 10083 10084 10085 10086 10087 10088 10089 10090 10091 10092 10093 10094 10095 10096 10097 10098 10099 10100


@templatedoc()
def get_tensor_from_selected_rows(x, name=None):
    """
    ${comment}

    Args:
        x(${x_type}): ${x_comment}
        name(basestring|None): Name of the output.

    Returns:
        out(${out_type}): ${out_comment}
    """

    helper = LayerHelper('get_tensor_from_selected_rows', **locals())
    out = helper.create_variable_for_type_inference(dtype=x.dtype)
    helper.append_op(
        type='get_tensor_from_selected_rows',
        inputs={'X': x},
        outputs={'Out': out},
        attrs={})
    return out
10101 10102


S
shippingwang 已提交
10103
def shuffle_channel(x, group, name=None):
S
shippingwang 已提交
10104 10105
    """
    **Shuffle Channel Operator**
10106

S
shippingwang 已提交
10107 10108 10109 10110 10111 10112
    This operator shuffles the channels of input x.
    It divide the input channels in each group into :attr:`group` subgroups,
    and obtain a new order by selecting element from every subgroup one by one.

    Please refer to the paper
    https://arxiv.org/pdf/1707.01083.pdf
S
shippingwang 已提交
10113
    
S
shippingwang 已提交
10114
    .. code-block:: text
10115

S
shippingwang 已提交
10116 10117 10118 10119 10120 10121 10122 10123 10124 10125 10126 10127 10128 10129 10130 10131 10132 10133 10134 10135 10136 10137 10138 10139 10140 10141 10142 10143
        Given a 4-D tensor input with the shape (N, C, H, W):
            input.shape = (1, 4, 2, 2)
            input.data =[[[[0.1, 0.2],
                           [0.2, 0.3]],

                          [[0.3, 0.4],
                           [0.4, 0.5]],

                          [[0.5, 0.6],
                           [0.6, 0.7]],

                          [[0.7, 0.8],
                           [0.8, 0.9]]]]
            Given group: 2
            then we get a 4-D tensor out whth the same shape of input:
            out.shape = (1, 4, 2, 2)
            out.data = [[[[0.1, 0.2],
                          [0.2, 0.3]],
                          
                         [[0.5, 0.6],
                          [0.6, 0.7]],
                          
                         [[0.3, 0.4],
                          [0.4, 0.5]],
                          
                         [[0.7, 0.8],
                          [0.8, 0.9]]]]
                        
S
shippingwang 已提交
10144
    Args: 
S
shippingwang 已提交
10145 10146
        x(Variable): The input tensor variable. It should be a 4-D tensor with shape [N, C, H, W]
        group(int): Indicating the conuts of subgroups, It should divide the number of channels.
S
shippingwang 已提交
10147 10148

    Returns:
S
shippingwang 已提交
10149 10150
        out(Variable): the channels shuffling result is a tensor variable with the 
        same shape and same type as the input.
S
shippingwang 已提交
10151 10152

    Raises:
S
shippingwang 已提交
10153
        ValueError: If group is not an int type variable.
S
shippingwang 已提交
10154 10155 10156

    Examples:
        .. code-block:: python
10157 10158

            input = fluid.layers.data(name='input', shape=[4,2,2], dtype='float32')
S
shippingwang 已提交
10159
            out = fluid.layers.shuffle_channel(x=input, group=2)
S
shippingwang 已提交
10160 10161 10162
    """
    helper = LayerHelper("shuffle_channel", **locals())

S
shippingwang 已提交
10163
    out = helper.create_variable_for_type_inference(dtype=x.dtype)
S
shippingwang 已提交
10164 10165 10166 10167 10168 10169 10170 10171 10172

    if not isinstance(group, int):
        raise TypeError("group must be int type")

    helper.append_op(
        type="shuffle_channel",
        inputs={"X": x},
        outputs={"Out": out},
        attrs={"group": group})
S
shippingwang 已提交
10173
    return out
S
Add  
shippingwang 已提交
10174 10175


S
sneaxiy 已提交
10176
class PyFuncRegistry(object):
S
sneaxiy 已提交
10177 10178 10179
    _register_funcs = []

    def __init__(self, func):
S
sneaxiy 已提交
10180
        if func is None or not callable(func):
S
sneaxiy 已提交
10181 10182 10183
            raise TypeError('func must be a Python function')

        self._func = func
M
minqiyang 已提交
10184
        # find named args using reflection
S
sneaxiy 已提交
10185 10186 10187 10188 10189 10190 10191
        args = inspect.getargspec(self._func)
        if len(args[0]) == 0 and args[1] is None and args[2] is None:
            # Function with no inputs
            self._named_args = None
        else:
            self._named_args = args[0]
        self._id = core._append_python_callable_object_and_return_id(self)
S
sneaxiy 已提交
10192 10193 10194
        '''
        Why record self here?

M
minqiyang 已提交
10195 10196
        1. For debug usage. Users can call
           :code:`py_func.registered_func(idx)` method
S
sneaxiy 已提交
10197
           to find the registered function corresponding
M
minqiyang 已提交
10198
           to :code:`idx`.
S
sneaxiy 已提交
10199

M
minqiyang 已提交
10200 10201
        2. For increasing reference count of self.
           It seems that to release Python object
S
sneaxiy 已提交
10202
           whose reference count is 1 would cause
M
minqiyang 已提交
10203
           segmentation fault error in C++ side.
S
sneaxiy 已提交
10204 10205
           May be lack of Python GC in C++ side?
        '''
S
sneaxiy 已提交
10206
        PyFuncRegistry._register_funcs.append(self)
S
sneaxiy 已提交
10207 10208 10209 10210 10211 10212 10213 10214 10215 10216 10217 10218 10219 10220

    @classmethod
    def registered_func(cls, idx):
        return cls._register_funcs[idx]._func

    @classmethod
    def registered_func_num(cls):
        return len(cls._register_funcs)

    @property
    def id(self):
        return self._id

    def __call__(self, *args):
S
sneaxiy 已提交
10221 10222 10223 10224 10225 10226 10227 10228 10229
        if self._named_args is None:
            func_ret = self._func()
        else:
            kwargs = dict()
            idx = 0
            for arg in self._named_args:
                kwargs[arg] = args[idx]
                idx += 1
            func_ret = self._func(*args[idx:], **kwargs)
S
sneaxiy 已提交
10230

S
sneaxiy 已提交
10231 10232
        if not isinstance(func_ret, (list, tuple)):
            func_ret = (func_ret, )
S
sneaxiy 已提交
10233 10234

        ret = []
S
sneaxiy 已提交
10235 10236 10237
        for each_ret in func_ret:
            if each_ret is None or isinstance(each_ret, core.LoDTensor):
                ret.append(each_ret)
S
sneaxiy 已提交
10238 10239
                continue

S
sneaxiy 已提交
10240 10241
            if not isinstance(each_ret, np.ndarray):
                each_ret = np.array(each_ret)
S
sneaxiy 已提交
10242

S
sneaxiy 已提交
10243 10244 10245
            tensor = core.LoDTensor()
            tensor.set(each_ret, core.CPUPlace())
            ret.append(tensor)
S
sneaxiy 已提交
10246

S
sneaxiy 已提交
10247
        return tuple(ret)
S
sneaxiy 已提交
10248 10249


S
sneaxiy 已提交
10250 10251 10252 10253
@templatedoc()
def py_func(func, x, out, backward_func=None, skip_vars_in_backward_input=None):
    """
    PyFunc Operator.
M
minqiyang 已提交
10254

S
sneaxiy 已提交
10255 10256 10257 10258 10259 10260 10261 10262
    User can use :code:`py_func` to register operators in Python side.
    The inputs of :code:`func` is :code:`LoDTensor` and outputs can be
    numpy array or :code:`LoDTensor`. Paddle would call the registered
    :code:`func` in forward part, and call :code:`backward_func` in
    backward part (if :code:`backward_func` is not None).

    User should set the right data type and shape of :code:`out` before
    calling this function. However, data types and shapes of gradients of
S
sneaxiy 已提交
10263
    :code:`out` and :code:`x` would be inferred automatically.
S
sneaxiy 已提交
10264

S
sneaxiy 已提交
10265 10266
    Input orders of :code:`backward_func` would be: forward inputs
    :code:`x`, forward outputs :code:`out` and backward input gradients of
S
sneaxiy 已提交
10267 10268 10269 10270
    :code:`out`. If some variables of :code:`out` have no gradient, the input
    tensor would be None in Python side. If some variables of :code:`in` have
    no gradient, users should return None.

S
sneaxiy 已提交
10271
    This function can also be used to debug the running network. User can
M
minqiyang 已提交
10272
    add a :code:`py_func` operator without output, and print input
S
sneaxiy 已提交
10273 10274
    :code:`x` inside :code:`func`.

S
sneaxiy 已提交
10275 10276 10277 10278 10279
    Args:
        func (callable): forward Python function.
        x (Variable|list(Variable)|tuple(Variable)): inputs of :code:`func`.
        out (Variable|list(Variable)|tuple(Variable)): outputs of :code:`func`.
            Paddle cannot infer shapes and data types of :code:`out`. Users
M
minqiyang 已提交
10280
            should create :code:`out` beforehand.
S
sneaxiy 已提交
10281
        backward_func (callable|None): backward Python function.
M
minqiyang 已提交
10282
                                       None means no backward. Default None.
S
sneaxiy 已提交
10283
        skip_vars_in_backward_input (Variable|list(Variable)|tuple(Variable)):
M
minqiyang 已提交
10284
            Variables that are not needed in :code:`backward_func` inputs.
S
sneaxiy 已提交
10285 10286
            These variables must be any of :code:`x` and :code:`out`.
            If set, these vars would not be inputs of :code:`backward_func`,
M
minqiyang 已提交
10287
            Only useful when :code:`backward_func` is not None. Default None.
S
sneaxiy 已提交
10288 10289 10290

    Returns:
        out (Variable|list(Variable)|tuple(Variable)): input :code:`out`
S
sneaxiy 已提交
10291 10292

    Examples:
M
minqiyang 已提交
10293

S
sneaxiy 已提交
10294 10295 10296 10297 10298
        >>> import paddle.fluid as fluid
        >>> import six
        >>>
        >>> def create_tmp_var(name, dtype, shape):
        >>>     return fluid.default_main_program().current_block().create_var(
M
minqiyang 已提交
10299
        >>>         name=name, dtype=dtype, shape=shape)
S
sneaxiy 已提交
10300 10301
        >>>
        >>> # tanh activation has been provided by Paddle C++ op
M
minqiyang 已提交
10302
        >>> # Here, we only use tanh to be an example to show the usage
S
sneaxiy 已提交
10303 10304 10305
        >>> # of py_func
        >>> def tanh(x):
        >>>     return np.tanh(x)
M
minqiyang 已提交
10306
        >>>
S
sneaxiy 已提交
10307 10308 10309 10310 10311
        >>> # forward input x is skipped
        >>> def tanh_grad(y, dy):
        >>>     return np.array(dy) * (1 - np.square(np.array(y)))
        >>>
        >>> def debug_func(x):
M
minqiyang 已提交
10312
        >>>     print(x)
S
sneaxiy 已提交
10313 10314 10315 10316 10317 10318
        >>>
        >>> def simple_net(img, label):
        >>>     hidden = img
        >>>     for idx in six.moves.range(4):
        >>>         hidden = fluid.layers.fc(hidden, size=200)
        >>>         new_hidden = create_tmp_var(name='hidden_{}'.format(idx),
M
minqiyang 已提交
10319
        >>>             dtype=hidden.dtype, shape=hidden.shape)
S
sneaxiy 已提交
10320 10321
        >>>
        >>>         # user-defined layers with forward and backward
M
minqiyang 已提交
10322 10323
        >>>         hidden = fluid.layers.py_func(func=tanh, x=hidden,
        >>>             out=new_hidden, backward_func=tanh_grad,
S
sneaxiy 已提交
10324 10325 10326 10327 10328 10329 10330 10331
        >>>             skip_vars_in_backward_input=hidden)
        >>>
        >>>         # user-defined debug layers to print variables
        >>>         fluid.layers.py_func(func=debug_func, x=hidden, out=None)
        >>>
        >>>     prediction = fluid.layers.fc(hidden, size=10, act='softmax')
        >>>     loss = fluid.layers.cross_entropy(input=prediction, label=label)
        >>>     return fluid.layers.mean(loss)
S
sneaxiy 已提交
10332
    """
S
sneaxiy 已提交
10333
    helper = LayerHelper('py_func', **locals())
S
sneaxiy 已提交
10334 10335 10336
    if x is None:
        x = []
    elif isinstance(x, Variable):
S
sneaxiy 已提交
10337
        x = [x]
S
sneaxiy 已提交
10338 10339
    elif not isinstance(x, (list, tuple)):
        raise TypeError('Input must be Variable/list(Variable)/tuple(Variable)')
S
sneaxiy 已提交
10340

S
sneaxiy 已提交
10341 10342 10343
    if out is None:
        out_list = []
    elif isinstance(out, Variable):
S
sneaxiy 已提交
10344
        out_list = [out]
S
sneaxiy 已提交
10345
    elif isinstance(out, (list, tuple)):
S
sneaxiy 已提交
10346
        out_list = out
S
sneaxiy 已提交
10347 10348 10349
    else:
        raise TypeError(
            'Output must be Variable/list(Variable)/tuple(Variable)')
S
sneaxiy 已提交
10350

S
sneaxiy 已提交
10351 10352
    fwd_func_id = PyFuncRegistry(func).id
    bwd_func_id = PyFuncRegistry(
S
sneaxiy 已提交
10353
        backward_func).id if backward_func is not None else -1
S
sneaxiy 已提交
10354 10355

    for each_out in out_list:
S
sneaxiy 已提交
10356 10357
        if len(each_out.shape) == 0:
            raise ValueError(
S
sneaxiy 已提交
10358 10359
                'Output shapes of py_func op should be provided by users manually'
            )
S
sneaxiy 已提交
10360

S
sneaxiy 已提交
10361 10362 10363 10364 10365 10366 10367 10368 10369 10370 10371 10372 10373 10374 10375
    backward_skip_vars = set()
    if backward_func is not None and skip_vars_in_backward_input is not None:
        if isinstance(skip_vars_in_backward_input, Variable):
            skip_vars_in_backward_input = [skip_vars_in_backward_input]

        fwd_in_out = [v.name for v in x]
        fwd_in_out.extend([v.name for v in out_list])
        fwd_in_out = set(fwd_in_out)
        backward_skip_vars = set()
        for v in skip_vars_in_backward_input:
            if not v.name in fwd_in_out:
                raise ValueError(
                    'Variable {} is not found in forward inputs and outputs'
                    .format(v.name))
            backward_skip_vars.add(v.name)
S
sneaxiy 已提交
10376 10377 10378 10379

    helper.append_op(
        type='py_func',
        inputs={'X': x},
S
sneaxiy 已提交
10380 10381
        outputs={'Out': out_list},
        attrs={
S
sneaxiy 已提交
10382 10383 10384
            'forward_callable_id': fwd_func_id,
            'backward_callable_id': bwd_func_id,
            'backward_skip_vars': list(backward_skip_vars)
S
sneaxiy 已提交
10385
        })
S
sneaxiy 已提交
10386
    return out
S
sneaxiy 已提交
10387 10388 10389


# For debug usage
S
sneaxiy 已提交
10390 10391 10392 10393
py_func.registered_func = PyFuncRegistry.registered_func
py_func.registered_func_num = PyFuncRegistry.registered_func_num


10394 10395 10396 10397 10398 10399 10400 10401 10402 10403 10404 10405 10406 10407 10408 10409 10410 10411 10412 10413 10414 10415 10416 10417 10418 10419 10420 10421 10422 10423 10424 10425 10426 10427 10428 10429 10430 10431 10432 10433 10434 10435 10436 10437 10438 10439 10440 10441 10442 10443 10444 10445
@templatedoc()
def psroi_pool(input,
               rois,
               output_channels,
               spatial_scale,
               pooled_height,
               pooled_width,
               name=None):
    """
    ${comment}

    Args:
        input (Variable): ${x_comment}
        rois (Variable): ROIs (Regions of Interest) to pool over.
        output_channels (integer): ${output_channels_comment}
        spatial_scale (float): ${spatial_scale_comment} Default: 1.0
        pooled_height (integer): ${pooled_height_comment} Default: 1
        pooled_width (integer): ${pooled_width_comment} Default: 1
        name (str, default None): The name of this layer.

    Returns:
        Variable: ${out_comment}.

    Examples:
        .. code-block:: python

            pool_out = fluid.layers.psroi_pool(input=x, rois=rois, 490, 1.0, 7, 7)
    """
    helper = LayerHelper('psroi_pool', **locals())
    # check attrs
    if not isinstance(output_channels, int):
        raise TypeError("output_channels must be int type")
    if not isinstance(spatial_scale, float):
        raise TypeError("spatial_scale must be float type")
    if not isinstance(pooled_height, int):
        raise TypeError("pooled_height must be int type")
    if not isinstance(pooled_width, int):
        raise TypeError("pooled_width must be int type")
    dtype = helper.input_dtype()
    out = helper.create_variable_for_type_inference(dtype)
    helper.append_op(
        type='psroi_pool',
        inputs={'X': input,
                'ROIs': rois},
        outputs={'Out': out},
        attrs={
            'output_channels': output_channels,
            'spatial_scale': spatial_scale,
            'pooled_height': pooled_height,
            'pooled_width': pooled_width
        })
    return out
10446

M
minqiyang 已提交
10447

M
minqiyang 已提交
10448
def huber_loss(input, label, delta):
10449
    """
M
minqiyang 已提交
10450 10451 10452
    Huber loss is a loss function used in robust.
    Huber loss can evaluate the fitness of input to label.
    Different from MSE loss, Huber loss is more robust for outliers.
10453 10454 10455 10456

    When the difference between input and label is large than delta
    .. math::

M
minqiyang 已提交
10457
        huber\_loss = delta * (label - input) - 0.5 * delta * delta
10458 10459 10460 10461

    When the difference between input and label is less than delta
    .. math::

M
minqiyang 已提交
10462
        huber\_loss = 0.5 * (label - input) * (label - input)
10463 10464 10465 10466 10467 10468 10469


    Args:
        input (Variable): This input is a probability computed by the previous operator.
                          The first dimension is batch size, and the last dimension is 1.
        label (Variable): The groud truth whose first dimension is batch size
                          and last dimension is 1.
M
minqiyang 已提交
10470
        delta (float): The parameter of huber loss, which controls
10471 10472 10473
                       the range of outliers

    Returns:
M
minqiyang 已提交
10474
        huber\_loss (Variable): The huber loss with shape [batch_size, 1].
10475 10476 10477 10478 10479

    Examples:
        .. code-block:: python

            predictions = fluid.layers.softmax(x)
M
minqiyang 已提交
10480
            loss = fluid.layers.huber_loss(input=predictions, label=label, 1.0)
10481
    """
M
minqiyang 已提交
10482
    helper = LayerHelper('huber_loss', **locals())
10483 10484 10485 10486 10487 10488 10489 10490 10491 10492 10493
    residual = helper.create_variable_for_type_inference(
        dtype=helper.input_dtype())
    out = helper.create_variable_for_type_inference(dtype=helper.input_dtype())
    helper.append_op(
        type='huber_loss',
        inputs={'X': input,
                'Y': label},
        outputs={'Out': out,
                 'Residual': residual},
        attrs={'delta': delta})
    return out
Z
zhaozhehao 已提交
10494 10495 10496 10497 10498 10499 10500 10501 10502 10503 10504 10505 10506 10507 10508 10509 10510 10511 10512 10513 10514 10515 10516 10517 10518 10519 10520 10521 10522 10523 10524 10525 10526 10527 10528 10529 10530 10531 10532 10533 10534 10535 10536 10537 10538 10539 10540 10541 10542 10543 10544 10545 10546 10547 10548 10549 10550 10551 10552 10553 10554 10555 10556 10557 10558 10559 10560 10561 10562 10563


@templatedoc()
def tree_conv(nodes_vector,
              edge_set,
              output_size,
              num_filters=1,
              max_depth=2,
              act='tanh',
              param_attr=None,
              bias_attr=None,
              name=None):
    """ 
    ${comment}
    		
    Args:
        nodes_vector(${nodes_vector_type}): ${nodes_vector_comment}
        edge_set(${edge_set_type}): ${edge_set_comment}
        output_size(int): output feature width
        num_filters(int): number of filters, Default 1
        max_depth(int): max depth of filters, Default 2
        act(str): activation function, Default tanh
        param_attr(ParamAttr): the parameter attribute for the filters, Default None
        bias_attr(ParamAttr): the parameter attribute for the bias of this layer, Default None
        name(str): a name of this layer(optional). If set None, the layer will be named automatically, Default None

    Returns:
        out(${out_type}): ${out_comment}

    Examples:
        .. code-block:: python

          nodes_vector = layers.data(name='vectors', shape=[None, 10, 5], dtype='float32)
          # None for batch size, 10 for max_node_size of dataset, 5 for vector width
          edge_set = layers.data(name='edge_set', shape=[None, 10, 2], dtype='float32')
          # None for batch size, 10 for max_node_size of dataset, 2 for every edge has two nodes
          # edges must be directional
          out_vector = layers.tree_conv(nodes_vector, edge_set, 6, 1, 2, 'tanh',
              ParamAttr(initializer=Constant(1.0), ParamAttr(initializer=Constant(1.0))
          # the shape of output will be [None, 10, 6, 1],
          # None for batch size, 10 for max_node_size of dataset, 6 for output size, 1 for 1 filter
          out_vector = layers.reshape(out_vector, shape=[None, 10, 6])
          # After reshape, output tensor could be nodes_vector for next tree convolution
          out_vector_2 = layers.tree_conv(out_vector, edge_set, 3, 4, 2, 'tanh',
              ParamAttr(initializer=Constant(1.0), ParamAttr(initializer=Constant(1.0))
          # also output tensor could be pooling(the pooling in paper called global pooling)
          pooled = layers.reduce_max(out_vector, dims=2) # global pooling
    """
    helper = LayerHelper("tree_conv", **locals())
    dtype = helper.input_dtype('nodes_vector')
    feature_size = nodes_vector.shape[2]
    W_shape = [feature_size, 3, output_size, num_filters]
    W = helper.create_parameter(
        attr=param_attr, shape=W_shape, dtype=dtype, is_bias=False)
    if name == None:
        out = helper.create_variable_for_type_inference(dtype=dtype)
    else:
        out = helper.create_variable(name=name, dtype=dtype, persistable=False)
    helper.append_op(
        type='tree_conv',
        inputs={'NodesVector': nodes_vector,
                'EdgeSet': edge_set,
                'Filter': W},
        outputs={'Out': out, },
        attrs={'max_depth': max_depth})
    if helper.bias_attr:
        pre_activation = helper.append_bias_op(out)
    else:
        pre_activation = out
    return helper.append_activation(pre_activation)
C
ceci3 已提交
10564 10565 10566 10567 10568 10569


def npair_loss(anchor, positive, labels, l2_reg=0.002):
    '''
  **Npair Loss Layer**
  
C
ceci3 已提交
10570 10571 10572 10573
  Read `Improved Deep Metric Learning with Multi class N pair Loss Objective <http://www.nec-labs.com/uploads/images/Department-Images/MediaAnalytics/papers/nips16_npairmetriclearning.pdf>`_ .
  
  Npair loss requires paired data. Npair loss has two parts: the first part is L2 
  regularizer on the embedding vector; the second part is cross entropy loss which
C
ceci3 已提交
10574 10575 10576 10577 10578
  takes the similarity matrix of anchor and positive as logits.

  Args:
    anchor(Variable): embedding vector for the anchor image. shape=[batch_size, embedding_dims]
    positive(Variable): embedding vector for the positive image. shape=[batch_size, embedding_dims]
C
ceci3 已提交
10579 10580
    labels(Variable): 1-D tensor. shape=[batch_size]
    l2_reg(float32): L2 regularization term on embedding vector, default: 0.002
C
ceci3 已提交
10581 10582 10583 10584 10585 10586 10587 10588 10589 10590 10591 10592 10593 10594 10595 10596 10597 10598 10599 10600 10601 10602 10603 10604 10605 10606 10607 10608 10609 10610 10611 10612

  Returns:
    npair loss(Variable): return npair loss, shape=[1]

  Examples:
    .. code-block:: python

       npair_loss = fluid.layers.npair_loss(anchor, positive, labels, l2_reg)
  '''
    Beta = 0.25
    batch_size = labels.shape[0]

    labels = reshape(labels, shape=[batch_size, 1], inplace=True)
    labels = expand(labels, expand_times=[1, batch_size])

    from .control_flow import equal
    from .ops import square

    labels = equal(labels, transpose(labels, perm=[1, 0])).astype('float32')
    labels = labels / reduce_sum(labels, dim=1, keep_dim=True)

    l2loss = reduce_mean(reduce_sum(square(anchor), 1)) \
             + reduce_mean(reduce_sum(square(positive), 1))
    l2loss = l2loss * Beta * l2_reg

    similarity_matrix = matmul(
        anchor, positive, transpose_x=False, transpose_y=True)
    softmax_value = softmax(similarity_matrix)
    cross_entropy = -1 * reduce_sum(labels * log(softmax_value), 0)
    celoss = reduce_mean(cross_entropy)

    return l2loss + celoss