layers.py 190.9 KB
Newer Older
1
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
Z
zhangjinchao01 已提交
2 3 4 5 6 7 8 9 10 11 12 13 14 15
#
# 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.

import functools
16
import collections
Y
Yu Yang 已提交
17
import inspect
Z
zhangjinchao01 已提交
18 19 20

from paddle.trainer.config_parser import *
from .activations import LinearActivation, SigmoidActivation, TanhActivation, \
Y
Yu Yang 已提交
21
    ReluActivation, IdentityActivation, SoftmaxActivation, BaseActivation
Z
zhangjinchao01 已提交
22 23 24 25
from .evaluators import *
from .poolings import MaxPooling, AvgPooling, BasePoolingType
from .attrs import *
from .default_decorators import *
26

Z
zhangjinchao01 已提交
27 28 29 30 31 32
try:
    import cPickle as pickle
except ImportError:
    import pickle
import copy

Q
qijun 已提交
33
__all__ = [
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56
    'full_matrix_projection',
    'AggregateLevel',
    'ExpandLevel',
    'identity_projection',
    'dotmul_projection',
    'dotmul_operator',
    'repeat_layer',
    'seq_reshape_layer',
    'table_projection',
    'mixed_layer',
    'data_layer',
    'embedding_layer',
    'fc_layer',
    'grumemory',
    'pooling_layer',
    'lstmemory',
    'last_seq',
    'first_seq',
    'cos_sim',
    'hsigmoid',
    'conv_projection',
    'mse_cost',
    'regression_cost',
Q
qijun 已提交
57
    'classification_cost',
58
    'LayerOutput',
Q
qijun 已提交
59 60 61 62 63 64
    'img_conv_layer',
    'img_pool_layer',
    'batch_norm_layer',
    'img_cmrnorm_layer',
    'addto_layer',
    'concat_layer',
65
    'seq_concat_layer',
Q
qijun 已提交
66 67 68 69 70 71
    'lstm_step_layer',
    'recurrent_group',
    'memory',
    'StaticInput',
    'expand_layer',
    'scaling_layer',
X
xuwei06 已提交
72
    'scaling_projection',
Q
qijun 已提交
73 74 75 76
    'power_layer',
    'interpolation_layer',
    'bilinear_interp_layer',
    'trans_layer',
77
    'rotate_layer',
Q
qijun 已提交
78 79 80 81 82 83 84 85 86
    'sum_to_one_norm_layer',
    'get_output_layer',
    'LayerType',
    'context_projection',
    'beam_search',
    'maxid_layer',
    'GeneratedInput',
    'SubsequenceInput',
    'gru_step_layer',
Y
Yu Yang 已提交
87
    'gru_step_naive_layer',
Q
qijun 已提交
88 89 90 91 92 93 94 95 96 97 98 99
    'recurrent_layer',
    'BaseGeneratedInput',
    'conv_operator',
    'conv_shift_layer',
    'tensor_layer',
    'selective_fc_layer',
    'sampling_id_layer',
    'slope_intercept_layer',
    'trans_full_matrix_projection',
    'linear_comb_layer',
    'convex_comb_layer',
    'ctc_layer',
100
    'warp_ctc_layer',
Q
qijun 已提交
101 102 103 104 105 106 107 108 109 110 111 112 113
    'crf_layer',
    'crf_decoding_layer',
    'nce_layer',
    'cross_entropy_with_selfnorm',
    'cross_entropy',
    'multi_binary_label_cross_entropy',
    'sum_cost',
    'rank_cost',
    'lambda_cost',
    'huber_cost',
    'block_expand_layer',
    'maxout_layer',
    'out_prod_layer',
X
xuwei06 已提交
114
    'printer_layer',
Q
qijun 已提交
115
    'print_layer',
Y
yuan 已提交
116
    'priorbox_layer',
117
    'cross_channel_norm_layer',
118 119
    'multibox_loss_layer',
    'detection_output_layer',
Q
qijun 已提交
120
    'spp_layer',
D
dangqingqing 已提交
121
    'pad_layer',
L
Luo Tao 已提交
122
    'eos_layer',
123
    'smooth_l1_cost',
124
    'layer_support',
W
wwhu 已提交
125
    'multiplex_layer',
D
dangqingqing 已提交
126
    'row_conv_layer',
127
    'dropout_layer',
128
    'prelu_layer',
129
    'gated_unit_layer',
Q
qijun 已提交
130
]
Z
zhangjinchao01 已提交
131 132 133 134 135 136 137


class LayerType(object):
    """
    Layer type enumerations.
    """

138 139 140 141 142 143 144 145
    DATA = 'data'
    MIXED_LAYER = 'mixed'
    LSTMEMORY = 'lstmemory'
    GRUMEMORY = 'gated_recurrent'
    SEQUENCE_LAST_INSTANCE = 'seqlastins'
    SEQUENCE_FIRST_INSTANCE = 'seqfirstins'
    SEQUENCE_RESHAPE = 'seqreshape'
    POOLING_MAX = 'max'
Z
zhangjinchao01 已提交
146
    POOLING_AVG = 'average'
147
    FC_LAYER = 'fc'
Z
zhangjinchao01 已提交
148
    COST = 'cost'
149 150
    COSINE_SIM_VEC = 'cos_vm'
    COSINE_SIM = 'cos'
Z
zhangjinchao01 已提交
151
    HSIGMOID = 'hsigmoid'
152 153 154 155 156 157
    CONV_LAYER = 'conv'
    CONVTRANS_LAYER = 'convt'
    EXCONV_LAYER = 'exconv'
    EXCONVTRANS_LAYER = 'exconvt'
    CUDNNCONV_LAYER = 'cudnn_conv'
    POOL_LAYER = 'pool'
Z
zhangjinchao01 已提交
158 159 160 161 162 163 164
    BATCH_NORM_LAYER = 'batch_norm'
    NORM_LAYER = 'norm'
    SUM_TO_ONE_NORM_LAYER = 'sum_to_one_norm'
    ADDTO_LAYER = 'addto'

    CONCAT_LAYER = 'concat'
    CONCAT_PROJ_LAYER = 'concat2'
165
    SEQUENCE_CONCAT_LAYER = 'seqconcat'
Z
zhangjinchao01 已提交
166 167 168 169 170 171 172

    LSTM_STEP_LAYER = 'lstm_step'
    GRU_STEP_LAYER = 'gru_step'
    GET_OUTPUT_LAYER = 'get_output'

    EXPAND_LAYER = 'expand'
    INTERPOLATION_LAYER = 'interpolation'
L
liaogang 已提交
173
    BILINEAR_INTERP_LAYER = 'bilinear_interp'
Z
zhangjinchao01 已提交
174 175 176
    POWER_LAYER = 'power'
    SCALING_LAYER = 'scaling'
    TRANS_LAYER = 'trans'
177
    ROTATE_LAYER = 'rotate'
H
Haonan 已提交
178
    OUT_PROD_LAYER = 'out_prod'
X
xuwei06 已提交
179
    FEATURE_MAP_EXPAND_LAYER = 'featmap_expand'
Z
zhangjinchao01 已提交
180 181 182 183 184 185 186 187 188 189 190

    MEMORY = 'memory'
    MAXID_LAYER = 'maxid'
    EOSID_LAYER = 'eos_id'
    RECURRENT_LAYER = 'recurrent'

    CONV_SHIFT_LAYER = "conv_shift"
    TENSOR_LAYER = "tensor"
    SEL_FC_LAYER = "selective_fc"
    SAMPLING_ID_LAYER = "sampling_id"
    SLOPE_INTERCEPT_LAYER = "slope_intercept"
191
    LINEAR_COMBINATION_LAYER = "convex_comb"
Z
zhangjinchao01 已提交
192
    BLOCK_EXPAND = "blockexpand"
193
    MAXOUT = "maxout"
Q
qijun 已提交
194
    SPP_LAYER = "spp"
D
dangqingqing 已提交
195
    PAD_LAYER = "pad"
W
wwhu 已提交
196
    MULTIPLEX_LAYER = "multiplex"
D
dangqingqing 已提交
197
    ROW_CONV_LAYER = "row_conv"
D
dangqingqing 已提交
198 199 200

    PRINT_LAYER = 'print'
    PRIORBOX_LAYER = 'priorbox'
201 202
    MULTIBOX_LOSS_LAYER = 'multibox_loss'
    DETECTION_OUTPUT_LAYER = 'detection_output'
D
dangqingqing 已提交
203 204 205 206 207

    CTC_LAYER = 'ctc'
    WARP_CTC_LAYER = 'warp_ctc'
    CRF_LAYER = 'crf'
    CRF_DECODING_LAYER = 'crf_decoding'
208
    NCE_LAYER = 'nce'
Z
zhangjinchao01 已提交
209

210 211 212 213 214 215 216 217 218 219 220
    RANK_COST = 'rank-cost'
    LAMBDA_COST = 'lambda_cost'
    HUBER = 'huber'
    CROSS_ENTROPY = 'multi-class-cross-entropy'
    CROSS_ENTROPY_WITH_SELFNORM = 'multi_class_cross_entropy_with_selfnorm'
    SOFT_BIN_CLASS_CROSS_ENTROPY = 'soft_binary_class_cross_entropy'
    MULTI_BIN_LABEL_CROSS_ENTROPY = 'multi_binary_label_cross_entropy'
    SUM_COST = 'sum_cost'
    SMOOTH_L1 = 'smooth_l1'

    PRELU = 'prelu'
Z
zhangjinchao01 已提交
221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241

    @staticmethod
    def is_layer_type(type_name):
        """
        If type_name is a layer type.

        :param type_name: layer type name. Because layer type enumerations are
                          strings.
        :type type_name: basestring
        :return: True if is a layer_type
        :rtype: bool
        """
        for key in dir(LayerType):
            if key.isupper():
                att = getattr(LayerType, key)
                if isinstance(att, basestring) and type_name == att:
                    return True
        return False


class AggregateLevel(object):
242
    """
L
Luo Tao 已提交
243
    PaddlePaddle supports three sequence types:
244 245 246

    - :code:`SequenceType.NO_SEQUENCE` means the sample is not a sequence.
    - :code:`SequenceType.SEQUENCE` means the sample is a sequence.
L
Luo Tao 已提交
247 248
    - :code:`SequenceType.SUB_SEQUENCE` means the sample is a nested sequence,
      each timestep of which is also a sequence.
249

L
Luo Tao 已提交
250
    Accordingly, AggregateLevel supports two modes:
251

L
Luo Tao 已提交
252
    - :code:`AggregateLevel.TO_NO_SEQUENCE` means the aggregation acts on each
L
Luo Tao 已提交
253
      timestep of a sequence, both :code:`SUB_SEQUENCE` and :code:`SEQUENCE` will
254 255
      be aggregated to :code:`NO_SEQUENCE`.

L
Luo Tao 已提交
256
    - :code:`AggregateLevel.TO_SEQUENCE` means the aggregation acts on each
257 258 259
      sequence of a nested sequence, :code:`SUB_SEQUENCE` will be aggregated to
      :code:`SEQUENCE`.
    """
L
Luo Tao 已提交
260 261
    TO_NO_SEQUENCE = 'non-seq'
    TO_SEQUENCE = 'seq'
262 263 264
    # compatible with previous configuration
    EACH_TIMESTEP = TO_NO_SEQUENCE
    EACH_SEQUENCE = TO_SEQUENCE
Z
zhangjinchao01 已提交
265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286


class LayerOutput(object):
    """
    LayerOutput is output for layer function. It is used internally by several
    reasons.

    - Check layer connection make sense.

        - FC(Softmax) => Cost(MSE Error) is not good for example.

    - Tracking layer connection.

    - Pass to layer methods as input.

    :param name: Layer output name.
    :type name: basestring
    :param layer_type: Current Layer Type. One of LayerType enumeration.
    :type layer_type: basestring
    :param activation: Layer Activation.
    :type activation: BaseActivation.
    :param parents: Layer's parents.
287
    :type parents: list|tuple|collections.Sequence
Z
zhangjinchao01 已提交
288 289
    """

Q
qijun 已提交
290 291 292 293 294 295 296 297 298
    def __init__(self,
                 name,
                 layer_type,
                 parents=None,
                 activation=None,
                 num_filters=None,
                 img_norm_type=None,
                 size=None,
                 outputs=None,
299
                 reverse=None):
Z
zhangjinchao01 已提交
300 301
        assert isinstance(name, basestring)
        assert isinstance(layer_type, basestring)
X
xuwei06 已提交
302
        assert size is not None
Z
zhangjinchao01 已提交
303 304
        assert LayerType.is_layer_type(layer_type)
        self.name = name
X
xuwei06 已提交
305
        self.full_name = MakeLayerNameInSubmodel(name)
Z
zhangjinchao01 已提交
306
        self.layer_type = layer_type
307 308
        if parents is not None and type(parents) != list:
            parents = [parents]
Z
zhangjinchao01 已提交
309 310 311 312 313 314 315 316
        self.parents = [] if parents is None else parents
        self.activation = activation
        self.num_filters = num_filters
        self.img_norm_type = img_norm_type
        self.size = size
        if outputs is None:
            outputs = ['default']
        self.outputs = outputs
317
        self.reverse = reverse
Z
zhangjinchao01 已提交
318

319 320 321 322 323 324 325 326
    def set_input(self, input):
        """
        Set the input for a memory layer. Can only be used for memory layer
        """
        assert isinstance(input, LayerOutput)
        assert self.layer_type == LayerType.MEMORY
        SetMemoryInput(self.name, input.name)

Z
zhangjinchao01 已提交
327 328 329

ERROR_CLIPPING = 'error_clipping_threshold'
DROPOUT = 'drop_rate'
330
DEVICE = 'device'
Z
zhangjinchao01 已提交
331 332 333


def layer_support(*attrs):
334
    attrs_list = list(attrs)
335
    attrs_list.append(DEVICE)
Q
qijun 已提交
336

Z
zhangjinchao01 已提交
337 338 339
    def decorator(method):
        @functools.wraps(method)
        def wrapper(*args, **kwargs):
340
            for attr in attrs_list:
Z
zhangjinchao01 已提交
341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356
                for each in args:
                    if isinstance(each, ExtraLayerAttribute):
                        setattr(each, '_'.join(['can', attr]), True)
                for key in kwargs:
                    val = kwargs[key]
                    if isinstance(val, ExtraLayerAttribute):
                        setattr(val, '_'.join(['can', attr]), True)
            for each in args:
                if isinstance(each, ExtraLayerAttribute):
                    each.check(method.__name__)
            for key in kwargs:
                val = kwargs[key]
                if isinstance(val, ExtraLayerAttribute):
                    val.check(method.__name__)
            return method(*args, **kwargs)

Y
Yu Yang 已提交
357 358 359 360 361
        if hasattr(method, 'argspec'):
            wrapper.argspec = method.argspec
        else:
            wrapper.argspec = inspect.getargspec(method)

Z
zhangjinchao01 已提交
362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400
        return wrapper

    return decorator


@wrap_param_attr_default()
def full_matrix_projection(input, size=0, param_attr=None):
    """
    Full Matrix Projection. It performs full matrix multiplication.

    ..  math::
        out.row[i] += in.row[i] * weight

    There are two styles of usage.

    1. When used in mixed_layer like this, you can only set the input:

    .. code-block:: python

       with mixed_layer(size=100) as m:
           m += full_matrix_projection(input=layer)

    2. When used as an independant object like this, you must set the size:

    .. code-block:: python

       proj = full_matrix_projection(input=layer,
                                     size=100,
                                     param_attr=ParamAttr(name='_proj'))

    :param input: input layer
    :type input: LayerOutput
    :param size: The parameter size. Means the width of parameter.
    :type size: int
    :param param_attr: Parameter config, None if use default.
    :type param_attr: ParameterAttribute
    :return: A FullMatrixProjection Object.
    :rtype: FullMatrixProjection
    """
Q
qijun 已提交
401 402
    proj = FullMatrixProjection(
        input_layer_name=input.name, size=size, **param_attr.attr)
Z
zhangjinchao01 已提交
403 404 405 406
    proj.origin = input
    return proj


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
@wrap_param_attr_default()
def trans_full_matrix_projection(input, size=0, param_attr=None):
    """
    Different from full_matrix_projection, this projection performs matrix
    multiplication, using transpose of weight.

    ..  math::
        out.row[i] += in.row[i] * w^\mathrm{T}

    :math:`w^\mathrm{T}` means transpose of weight.
    The simply usage is:

    .. code-block:: python

       proj = trans_full_matrix_projection(input=layer,
                                           size=100,
                                           param_attr=ParamAttr(
                                                name='_proj',
                                                initial_mean=0.0,
                                                initial_std=0.01))

    :param input: input layer
    :type input: LayerOutput
    :param size: The parameter size. Means the width of parameter.
    :type size: int
    :param param_attr: Parameter config, None if use default.
    :type param_attr: ParameterAttribute
    :return: A TransposedFullMatrixProjection Object.
    :rtype: TransposedFullMatrixProjection
    """
Q
qijun 已提交
437 438
    proj = TransposedFullMatrixProjection(
        input_layer_name=input.name, size=size, **param_attr.attr)
439 440 441 442
    proj.origin = input
    return proj


Z
zhangjinchao01 已提交
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
@wrap_param_attr_default()
def table_projection(input, size=0, param_attr=None):
    """
    Table Projection. It selects rows from parameter where row\_id
    is in input\_ids.

    .. math::
       out.row[i] += table.row[ids[i]]

    where :math:`out` is output, :math:`table` is parameter, :math:`ids` is input\_ids,
    and :math:`i` is row\_id.

    There are two styles of usage.

    1. When used in mixed_layer like this, you can only set the input:

    .. code-block:: python

       with mixed_layer(size=100) as m:
           m += table_projection(input=layer)

    2. When used as an independant object like this, you must set the size:

    .. code-block:: python

       proj = table_projection(input=layer,
                               size=100,
                               param_attr=ParamAttr(name='_proj'))


    :param input: Input layer, which must contains id fields.
    :type input: LayerOutput
    :param size: The parameter size. Means the width of parameter.
    :type size: int
    :param param_attr: Parameter config, None if use default.
    :type param_attr: ParameterAttribute
    :return: A TableProjection Object.
    :rtype: TableProjection
    """
Q
qijun 已提交
482 483
    proj = TableProjection(
        input_layer_name=input.name, size=size, **param_attr.attr)
Z
zhangjinchao01 已提交
484 485 486 487
    proj.origin = input
    return proj


488
def identity_projection(input, offset=None, size=None):
Z
zhangjinchao01 已提交
489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518
    """
    1. IdentityProjection if offset=None. It performs:

    .. math::
       out.row[i] += in.row[i]

    The example usage is:

    .. code-block:: python

       proj = identity_projection(input=layer)


    2. IdentityOffsetProjection if offset!=None. It likes IdentityProjection,
    but layer size may be smaller than input size.
    It select dimesions [offset, offset+layer_size) from input:

    .. math::
       out.row[i] += in.row[i + \\textrm{offset}]

    The example usage is:

    .. code-block:: python

       proj = identity_projection(input=layer,
                                  offset=10)

    Note that both of two projections should not have any parameter.

    :param input: Input Layer.
519
    :type input: LayerOutput
Z
zhangjinchao01 已提交
520 521
    :param offset: Offset, None if use default.
    :type offset: int
X
xuwei06 已提交
522
    :return: A IdentityProjection or IdentityOffsetProjection object
Z
zhangjinchao01 已提交
523 524 525 526 527 528
    :rtype: IdentityProjection or IdentityOffsetProjection
    """
    if offset is None:
        proj = IdentityProjection(input_layer_name=input.name)
        proj.origin = input
    else:
529 530
        if size is None:
            size = input.size - offset
Q
qijun 已提交
531
        proj = IdentityOffsetProjection(
532
            input_layer_name=input.name, offset=offset, size=size)
Z
zhangjinchao01 已提交
533 534 535 536
        proj.origin = input
    return proj


X
xuwei06 已提交
537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558
@wrap_param_attr_default()
def scaling_projection(input, param_attr=None):
    """
    scaling_projection multiplies the input with a scalar parameter and add to
    the output.

    .. math::
       out += w * in

    The example usage is:

    .. code-block:: python

       proj = scaling_projection(input=layer)

    :param input: Input Layer.
    :type input: LayerOutput
    :param param_attr: Parameter config, None if use default.
    :type param_attr: ParameterAttribute
    :return: A ScalingProjection object
    :rtype: ScalingProjection
    """
L
Luo Tao 已提交
559
    proj = ScalingProjection(input_layer_name=input.name, **param_attr.attr)
X
xuwei06 已提交
560 561 562 563
    proj.origin = input
    return proj


Z
zhangjinchao01 已提交
564
@wrap_param_attr_default()
565
def dotmul_projection(input, param_attr=None):
Z
zhangjinchao01 已提交
566
    """
567
    DotMulProjection with a layer as input.
Z
zhangjinchao01 已提交
568 569 570 571 572 573 574 575 576 577 578 579 580
    It performs element-wise multiplication with weight.

    ..  math::
        out.row[i] += in.row[i] .* weight

    where :math:`.*` means element-wise multiplication.

    The example usage is:

    .. code-block:: python

       proj = dotmul_projection(input=layer)

581 582 583 584 585 586 587
    :param input: Input layer.
    :type input: LayerOutput
    :param param_attr: Parameter config, None if use default.
    :type param_attr: ParameterAttribute
    :return: A DotMulProjection Object.
    :rtype: DotMulProjection
    """
Q
qijun 已提交
588 589
    proj = DotMulProjection(
        input_layer_name=input.name, size=input.size, **param_attr.attr)
590
    proj.origin = input
591
    return proj
Z
zhangjinchao01 已提交
592

593 594

def dotmul_operator(a=None, b=None, scale=1, **kwargs):
595 596
    """
    DotMulOperator takes two inputs and performs element-wise multiplication:
597

Z
zhangjinchao01 已提交
598
    .. math::
L
Luo Tao 已提交
599
       out.row[i] += scale * (a.row[i] .* b.row[i])
600

Z
zhangjinchao01 已提交
601 602
    where :math:`.*` means element-wise multiplication, and
    scale is a config scalar, its default value is one.
603

Z
zhangjinchao01 已提交
604
    The example usage is:
605

Z
zhangjinchao01 已提交
606
    .. code-block:: python
607

L
Luo Tao 已提交
608
       op = dotmul_operator(a=layer1, b=layer2, scale=0.5)
609

610 611 612 613
    :param a: Input layer1
    :type a: LayerOutput
    :param b: Input layer2
    :type b: LayerOutput
Z
zhangjinchao01 已提交
614 615
    :param scale: config scalar, default value is one.
    :type scale: float
616 617
    :return: A DotMulOperator Object.
    :rtype: DotMulOperator
Z
zhangjinchao01 已提交
618
    """
619 620 621
    if 'x' in kwargs or 'y' in kwargs:
        logger.warning('x and y arguments for dotmul_operator is deprecated. '
                       'Please use a and b as parameter.')
Q
qijun 已提交
622
    a = kwargs.get('x', a)  # For Backward capacity.
623 624 625 626 627 628
    b = kwargs.get('y', b)
    assert isinstance(a, LayerOutput)
    assert isinstance(b, LayerOutput)
    if a.size is not None and b.size is not None:
        assert a.size == b.size

Q
qijun 已提交
629
    op = DotMulOperator(input_layer_names=[a.name, b.name], scale=scale)
630
    op.origin = [a, b]
631
    return op
Z
zhangjinchao01 已提交
632

633

Z
zhangjinchao01 已提交
634
@wrap_bias_attr_default(['padding_attr'])
Q
qijun 已提交
635 636 637
def context_projection(input,
                       context_len,
                       context_start=None,
Z
zhangjinchao01 已提交
638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673
                       padding_attr=False):
    """
    Context Projection.

    It just simply reorganizes input sequence, combines "context_len" sequence
    to one context from context_start. "context_start" will be set to
    -(context_len - 1) / 2 by default. If context position out of sequence
    length, padding will be filled as zero if padding_attr = False, otherwise
    it is trainable.

    For example, origin sequence is [A B C D E F G], context len is 3, then
    after context projection and not set padding_attr, sequence will
    be [ 0AB ABC BCD CDE DEF EFG FG0 ].

    :param input: Input Sequence.
    :type input: LayerOutput
    :param context_len: context length.
    :type context_len: int
    :param context_start: context start position. Default is
                          -(context_len - 1)/2
    :type context_start: int
    :param padding_attr: Padding Parameter Attribute. If false, it means padding
                         always be zero. Otherwise Padding is learnable, and
                         parameter attribute is set by this parameter.
    :type padding_attr: bool|ParameterAttribute
    :return: Projection
    :rtype: Projection
    """
    context_start = -(
        context_len - 1) / 2 if context_start is None else context_start

    extra_dict = dict()
    trainable = isinstance(padding_attr, ParameterAttribute)
    if trainable:
        extra_dict = padding_attr.attr

Q
qijun 已提交
674 675 676 677 678 679
    proj = ContextProjection(
        input_layer_name=input.name,
        context_length=context_len,
        context_start=context_start,
        trainable_padding=trainable,
        **extra_dict)
Z
zhangjinchao01 已提交
680 681 682 683 684 685 686 687 688 689 690 691 692
    proj.origin = input
    return proj


class MixedLayerType(LayerOutput):
    """
    The internal object for trainer_helpers.
    """

    class AddToSealedMixedLayerException(Exception):
        def __init__(self):
            Exception.__init__(self)

Q
qijun 已提交
693
    def __init__(self, name, size, act, bias_attr, layer_attr, parents=None):
Z
zhangjinchao01 已提交
694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709
        """
        Ctor.
        :param name: layer name.
        :type name: basestring
        :param size: layer size.
        :type size: int
        :param act: activation type.
        :type act: BaseActivation
        :param bias_attr: The Bias Attribute. If no bias, then pass False or
                          something not type of ParameterAttribute. None will
                          get a default Bias.
        :type bias_attr: ParameterAttribute or None means has bias. Any other
                         type means no bias.
        :param layer_attr: Extra Layer Attribute.
        :type layer_attr: ExtraLayerAttribute or None
        """
Q
qijun 已提交
710 711 712 713 714 715 716
        LayerOutput.__init__(
            self,
            name,
            LayerType.MIXED_LAYER,
            parents,
            size=size,
            activation=act)
Z
zhangjinchao01 已提交
717 718 719 720 721
        self.bias_attr = bias_attr
        self.layer_attr = layer_attr
        self.inputs = []
        self.finalized = False

722
    def __iadd__(self, other):
Z
zhangjinchao01 已提交
723 724 725 726 727 728 729 730
        """
        + += operator
        :param other: Other projection.
        :type other: Projection
        :return: self.
        :rtype: MixedLayerType
        """
        if not self.finalized:
731
            assert isinstance(other, Projection) or isinstance(other, Operator)
Z
zhangjinchao01 已提交
732
            self.inputs.append(other)
733 734 735 736
            if isinstance(other, Projection):
                self.parents.append(other.origin)
            else:
                self.parents.extend(other.origin)
Z
zhangjinchao01 已提交
737 738 739 740 741 742 743 744
            return self
        else:
            raise MixedLayerType.AddToSealedMixedLayerException()

    def __enter__(self):
        assert len(self.inputs) == 0
        return self

745
    def __exit__(self, exc_type, exc_value, tb):
W
wangyang59 已提交
746 747
        if exc_value is not None:
            raise exc_value
Z
zhangjinchao01 已提交
748
        assert len(self.inputs) != 0
749
        ml = MixedLayer(
Z
zhangjinchao01 已提交
750 751 752 753 754
            name=self.name,
            size=self.size,
            active_type=self.activation.name,
            bias=ParamAttr.to_bias(self.bias_attr),
            inputs=self.inputs,
Q
qijun 已提交
755
            **ExtraLayerAttribute.to_kwargs(self.layer_attr))
756 757 758
        # update the size which might be computed inside MixedLayer
        # according to the operator's output size
        self.size = ml.config.size
759
        self.finalized = True
Z
zhangjinchao01 已提交
760 761 762 763 764 765


@wrap_name_default("mixed")
@wrap_act_default(act=LinearActivation())
@wrap_bias_attr_default(has_bias=False)
@layer_support(ERROR_CLIPPING, DROPOUT)
Q
qijun 已提交
766 767 768 769 770
def mixed_layer(size=0,
                input=None,
                name=None,
                act=None,
                bias_attr=False,
Z
zhangjinchao01 已提交
771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814
                layer_attr=None):
    """
    Mixed Layer. A mixed layer will add all inputs together, then activate.
    Each inputs is a projection or operator.

    There are two styles of usages.

    1. When not set inputs parameter, use mixed_layer like this:

    .. code-block:: python

       with mixed_layer(size=256) as m:
           m += full_matrix_projection(input=layer1)
           m += identity_projection(input=layer2)

    2. You can also set all inputs when invoke mixed_layer as follows:

    .. code-block:: python

       m = mixed_layer(size=256,
                       input=[full_matrix_projection(input=layer1),
                              full_matrix_projection(input=layer2)])

    :param name: mixed layer name. Can be referenced by other layer.
    :type name: basestring
    :param size: layer size.
    :type size: int
    :param input: inputs layer. It is an optional parameter. If set,
                  then this function will just return layer's name.
    :param act: Activation Type.
    :type act: BaseActivation
    :param bias_attr: The Bias Attribute. If no bias, then pass False or
                      something not type of ParameterAttribute. None will get a
                      default Bias.
    :type bias_attr: ParameterAttribute or None or bool
    :param layer_attr: The extra layer config. Default is None.
    :type layer_attr: ExtraLayerAttribute
    :return: MixedLayerType object can add inputs or layer name.
    :rtype: MixedLayerType
    """

    if input is None:
        return MixedLayerType(name, size, act, bias_attr, layer_attr)
    else:
Q
qijun 已提交
815 816 817 818 819 820
        with mixed_layer(
                name=name,
                size=size,
                act=act,
                bias_attr=bias_attr,
                layer_attr=layer_attr) as m:
821
            if isinstance(input, collections.Sequence):
Z
zhangjinchao01 已提交
822 823 824 825 826 827 828 829
                for each in input:
                    m += each
            else:
                m += input
        return m


@layer_support()
L
Luo Tao 已提交
830
def data_layer(name, size, height=None, width=None, layer_attr=None):
Z
zhangjinchao01 已提交
831 832 833 834 835 836 837
    """
    Define DataLayer For NeuralNetwork.

    The example usage is:

    ..  code-block:: python

Y
Yu Yang 已提交
838
        data = data_layer(name="input", size=1000)
Z
zhangjinchao01 已提交
839 840 841 842 843

    :param name: Name of this data layer.
    :type name: basestring
    :param size: Size of this data layer.
    :type size: int
L
Luo Tao 已提交
844
    :param height: Height of this data layer, used for image
Y
Yu Yang 已提交
845
    :type height: int|None
L
Luo Tao 已提交
846
    :param width: Width of this data layer, used for image
Y
Yu Yang 已提交
847
    :type width: int|None
Z
zhangjinchao01 已提交
848 849
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute.
D
dangqingqing 已提交
850
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
851 852
    :rtype: LayerOutput
    """
Q
qijun 已提交
853 854 855 856
    Layer(
        type=LayerType.DATA,
        name=name,
        size=size,
L
Luo Tao 已提交
857 858
        height=height,
        width=width,
Q
qijun 已提交
859
        **ExtraLayerAttribute.to_kwargs(layer_attr))
Z
zhangjinchao01 已提交
860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881

    return LayerOutput(name, LayerType.DATA, size=size)


@wrap_name_default("embedding")
@wrap_param_attr_default()
@layer_support(ERROR_CLIPPING)
def embedding_layer(input, size, name=None, param_attr=None, layer_attr=None):
    """
    Define a embedding Layer.

    :param name: Name of this embedding layer.
    :type name: basestring
    :param input: The input layer for this embedding. NOTE: must be Index Data.
    :type input: LayerOutput
    :param size: The embedding dimension.
    :type size: int
    :param param_attr: The embedding parameter attribute. See ParameterAttribute
                      for details.
    :type param_attr: ParameterAttribute|None
    :param layer_attr: Extra layer Config. Default is None.
    :type layer_attr: ExtraLayerAttribute|None
D
dangqingqing 已提交
882
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
883 884
    :rtype: LayerOutput
    """
Q
qijun 已提交
885 886 887 888 889 890
    with mixed_layer(
            name=name,
            size=size,
            act=LinearActivation(),
            bias_attr=False,
            layer_attr=layer_attr) as mix:
Z
zhangjinchao01 已提交
891 892 893 894 895 896 897 898 899
        mix += table_projection(input=input, size=size, param_attr=param_attr)
    return mix


@wrap_name_default()
@wrap_param_attr_default()
@wrap_bias_attr_default()
@wrap_act_default()
@layer_support(ERROR_CLIPPING, DROPOUT)
Q
qijun 已提交
900 901 902 903 904 905 906
def fc_layer(input,
             size,
             act=None,
             name=None,
             param_attr=None,
             bias_attr=None,
             layer_attr=None):
Z
zhangjinchao01 已提交
907 908 909 910 911 912 913 914 915 916 917 918
    """
    Helper for declare fully connected layer.

    The example usage is:

    .. code-block:: python

       fc = fc_layer(input=layer,
                     size=1024,
                     act=LinearActivation(),
                     bias_attr=False)

L
luotao02 已提交
919
    which is equal to:
Z
zhangjinchao01 已提交
920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941

    .. code-block:: python

       with mixed_layer(size=1024) as fc:
           fc += full_matrix_projection(input=layer)

    :param name: The Layer Name.
    :type name: basestring
    :param input: The input layer. Could be a list/tuple of input layer.
    :type input: LayerOutput|list|tuple
    :param size: The layer dimension.
    :type size: int
    :param act: Activation Type. Default is tanh.
    :type act: BaseActivation
    :param param_attr: The Parameter Attribute|list.
    :type param_attr: ParameterAttribute
    :param bias_attr: The Bias Attribute. If no bias, then pass False or
                      something not type of ParameterAttribute. None will get a
                      default Bias.
    :type bias_attr: ParameterAttribute|None|Any
    :param layer_attr: Extra Layer config.
    :type layer_attr: ExtraLayerAttribute|None
D
dangqingqing 已提交
942
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
943 944 945 946
    :rtype: LayerOutput
    """
    if isinstance(input, LayerOutput):
        input = [input]
947
        assert not isinstance(param_attr, collections.Sequence)
Z
zhangjinchao01 已提交
948 949
        param_attr = [param_attr]
    else:
950
        if isinstance(param_attr, collections.Sequence):
Z
zhangjinchao01 已提交
951 952 953 954
            assert len(input) == len(param_attr)
        else:
            param_attr = [copy.deepcopy(param_attr) for _ in range(len(input))]

955
    assert isinstance(input, collections.Sequence)
Z
zhangjinchao01 已提交
956 957

    Layer(
Q
qijun 已提交
958 959 960
        inputs=[
            Input(ipt.name, **attr.attr) for ipt, attr in zip(input, param_attr)
        ],
Z
zhangjinchao01 已提交
961 962 963 964 965
        name=name,
        type=LayerType.FC_LAYER,
        size=size,
        bias=ParamAttr.to_bias(bias_attr),
        active_type=act.name,
Q
qijun 已提交
966 967 968
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name, LayerType.FC_LAYER, input, activation=act, size=size)
Z
zhangjinchao01 已提交
969

970

971
@wrap_name_default("print")
972
def printer_layer(input, format=None, name=None):
973 974
    """
    Print the output value of input layers. This layer is useful for debugging.
975 976 977 978 979

    :param name: The Layer Name.
    :type name: basestring
    :param input: The input layer. Could be a list/tuple of input layer.
    :type input: LayerOutput|list|tuple
980
    :return: LayerOutput
981
    """
982 983 984 985 986
    if isinstance(input, LayerOutput):
        input = [input]
    assert isinstance(input, collections.Sequence)  # list or tuple
    for each in input:
        assert isinstance(each, LayerOutput)
987 988 989

    Layer(
        name=name,
990
        format=format,
991
        type=LayerType.PRINT_LAYER,
Q
qijun 已提交
992
        inputs=[l.name for l in input], )
993
    # this layer don't return anything, can not be input of other layer.
994

X
xuwei06 已提交
995 996 997 998 999 1000 1001
# Keep print_layer for compatibility with V1 API.
# 'print_layer' does not work for V2 API because it will be changed to
# 'print' for V2 API. But 'print' is a reserved key word in python.


print_layer = printer_layer

Z
zhangjinchao01 已提交
1002

Y
yuan 已提交
1003
@wrap_name_default("priorbox")
G
gaoyuan 已提交
1004
def priorbox_layer(input,
G
gaoyuan 已提交
1005
                   image,
G
gaoyuan 已提交
1006 1007 1008 1009 1010
                   aspect_ratio,
                   variance,
                   min_size,
                   max_size=[],
                   name=None):
Y
yuan 已提交
1011 1012 1013 1014 1015 1016 1017
    """
    Compute the priorbox and set the variance. This layer is necessary for ssd.

    :param name: The Layer Name.
    :type name: basestring
    :param input: The input layer.
    :type input: LayerOutput
G
gaoyuan 已提交
1018 1019
    :param image: The network input image.
    :type image: LayerOutput
Y
yuan 已提交
1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030
    :param aspect_ratio: The aspect ratio.
    :type aspect_ratio: list
    :param variance: The bounding box variance.
    :type min_size: The min size of the priorbox width/height.
    :param min_size: list
    :type max_size: The max size of the priorbox width/height. Could be NULL.
    :param max_size: list
    :return: LayerOutput
    """
    # plus one for ratio 1.
    num_filters = (len(aspect_ratio) * 2 + 1 + len(max_size)) * 4
G
gaoyuan 已提交
1031
    size = (input.size / input.num_filters) * num_filters * 2
Y
yuan 已提交
1032 1033 1034
    Layer(
        name=name,
        type=LayerType.PRIORBOX_LAYER,
G
gaoyuan 已提交
1035
        inputs=[input.name, image.name],
Y
yuan 已提交
1036 1037 1038 1039 1040 1041
        size=size,
        min_size=min_size,
        max_size=max_size,
        aspect_ratio=aspect_ratio,
        variance=variance)
    return LayerOutput(
G
gaoyuan 已提交
1042 1043
        name,
        LayerType.PRIORBOX_LAYER,
G
gaoyuan 已提交
1044
        parents=[input, image],
G
gaoyuan 已提交
1045 1046 1047
        num_filters=num_filters,
        size=size)

Z
zhangjinchao01 已提交
1048

1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064
@wrap_name_default("multibox_loss")
def multibox_loss_layer(input_loc,
                        input_conf,
                        priorbox,
                        label,
                        num_classes,
                        overlap_threshold=0.5,
                        neg_pos_ratio=3.0,
                        neg_overlap=0.5,
                        background_id=0,
                        name=None):
    """
    Compute the location loss and the confidence loss for ssd.

    :param name: The Layer Name.
    :type name: basestring
Y
yangyaming 已提交
1065 1066
    :param input_loc: The input predict locations.
    :type input_loc: LayerOutput | List of LayerOutput
1067
    :param input_conf: The input priorbox confidence.
Y
yangyaming 已提交
1068
    :type input_conf: LayerOutput | List of LayerOutput
1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089
    :param priorbox: The input priorbox location and the variance.
    :type priorbox: LayerOutput
    :param label: The input label.
    :type label: LayerOutput
    :param num_classes: The number of the classification.
    :type num_classes: int
    :param overlap_threshold: The threshold of the overlap.
    :type overlap_threshold: float
    :param neg_pos_ratio: The ratio of the negative bbox to the positive bbox.
    :type neg_pos_ratio: float
    :param neg_overlap: The negative bbox overlap threshold.
    :type neg_overlap: float
    :param background_id: The background class index.
    :type background_id: int
    :return: LayerOutput
    """
    if isinstance(input_loc, LayerOutput):
        input_loc = [input_loc]
    assert isinstance(input_loc, collections.Sequence)  # list or tuple
    for each in input_loc:
        assert isinstance(each, LayerOutput)
1090
    input_loc_num = len(input_loc)
1091 1092 1093 1094 1095 1096

    if isinstance(input_conf, LayerOutput):
        input_conf = [input_conf]
    assert isinstance(input_conf, collections.Sequence)  # list or tuple
    for each in input_conf:
        assert isinstance(each, LayerOutput)
1097
    input_conf_num = len(input_conf)
1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138
    # Check the input layer number.
    assert input_loc_num == input_conf_num

    inputs = [priorbox.name, label.name]
    inputs.extend([l.name for l in input_loc])
    inputs.extend([l.name for l in input_conf])
    parents = [priorbox, label]
    parents.extend(input_loc)
    parents.extend(input_conf)

    Layer(
        name=name,
        type=LayerType.MULTIBOX_LOSS_LAYER,
        inputs=inputs,
        input_num=input_loc_num,
        num_classes=num_classes,
        overlap_threshold=overlap_threshold,
        neg_pos_ratio=neg_pos_ratio,
        neg_overlap=neg_overlap,
        background_id=background_id)
    return LayerOutput(
        name, LayerType.MULTIBOX_LOSS_LAYER, parents=parents, size=1)


@wrap_name_default("detection_output")
def detection_output_layer(input_loc,
                           input_conf,
                           priorbox,
                           num_classes,
                           nms_threshold=0.45,
                           nms_top_k=400,
                           keep_top_k=200,
                           confidence_threshold=0.01,
                           background_id=0,
                           name=None):
    """
    Apply the NMS to the output of network and compute the predict bounding
    box location.

    :param name: The Layer Name.
    :type name: basestring
Y
yangyaming 已提交
1139 1140
    :param input_loc: The input predict locations.
    :type input_loc: LayerOutput | List of LayerOutput.
1141
    :param input_conf: The input priorbox confidence.
Y
yangyaming 已提交
1142
    :type input_conf: LayerOutput | List of LayerOutput.
1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163
    :param priorbox: The input priorbox location and the variance.
    :type priorbox: LayerOutput
    :param num_classes: The number of the classification.
    :type num_classes: int
    :param nms_threshold: The Non-maximum suppression threshold.
    :type nms_threshold: float
    :param nms_top_k: The bbox number kept of the NMS's output
    :type nms_top_k: int
    :param keep_top_k: The bbox number kept of the layer's output
    :type keep_top_k: int
    :param confidence_threshold: The classification confidence threshold
    :type confidence_threshold: float
    :param background_id: The background class index.
    :type background_id: int
    :return: LayerOutput
    """
    if isinstance(input_loc, LayerOutput):
        input_loc = [input_loc]
    assert isinstance(input_loc, collections.Sequence)  # list or tuple
    for each in input_loc:
        assert isinstance(each, LayerOutput)
Y
yangyaming 已提交
1164
    input_loc_num = len(input_loc)
1165 1166 1167 1168 1169 1170

    if isinstance(input_conf, LayerOutput):
        input_conf = [input_conf]
    assert isinstance(input_conf, collections.Sequence)  # list or tuple
    for each in input_conf:
        assert isinstance(each, LayerOutput)
Y
yangyaming 已提交
1171 1172
    input_conf_num = len(input_conf)

1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200
    # Check the input layer number.
    assert input_loc_num == input_conf_num

    inputs = [priorbox.name]
    inputs.extend([l.name for l in input_loc])
    inputs.extend([l.name for l in input_conf])
    parents = [priorbox]
    parents.extend(input_loc)
    parents.extend(input_conf)

    size = keep_top_k * 7

    Layer(
        name=name,
        type=LayerType.DETECTION_OUTPUT_LAYER,
        inputs=inputs,
        size=size,
        input_num=input_loc_num,
        num_classes=num_classes,
        nms_threshold=nms_threshold,
        nms_top_k=nms_top_k,
        keep_top_k=keep_top_k,
        confidence_threshold=confidence_threshold,
        background_id=background_id)
    return LayerOutput(
        name, LayerType.DETECTION_OUTPUT_LAYER, parents=parents, size=size)


1201 1202
@wrap_name_default("cross_channel_norm")
def cross_channel_norm_layer(input, name=None, param_attr=None):
G
gaoyuan 已提交
1203 1204 1205 1206 1207
    """
    Normalize a layer's output. This layer is necessary for ssd.
    This layer applys normalize across the channels of each sample to
    a conv layer's output and scale the output by a group of trainable
    factors which dimensions equal to the channel's number.
G
gaoyuan 已提交
1208

G
gaoyuan 已提交
1209 1210 1211 1212 1213 1214 1215 1216
    :param name: The Layer Name.
    :type name: basestring
    :param input: The input layer.
    :type input: LayerOutput
    :param param_attr: The Parameter Attribute|list.
    :type param_attr: ParameterAttribute
    :return: LayerOutput
    """
1217
    assert input.num_filters is not None
G
gaoyuan 已提交
1218 1219
    Layer(
        name=name,
1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232
        type=LayerType.NORM_LAYER,
        inputs=[
            Input(
                input.name,
                norm=Norm(
                    norm_type="cross-channel-norm",
                    channels=input.num_filters,
                    size=input.size,
                    scale=0,
                    pow=0,
                    blocked=0),
                **param_attr.attr)
        ])
G
gaoyuan 已提交
1233 1234
    return LayerOutput(
        name,
1235
        LayerType.NORM_LAYER,
G
gaoyuan 已提交
1236 1237 1238 1239 1240
        parents=input,
        num_filters=input.num_filters,
        size=input.size)


Z
zhangjinchao01 已提交
1241 1242 1243 1244
@wrap_name_default("seq_pooling")
@wrap_bias_attr_default(has_bias=False)
@wrap_param_default(['pooling_type'], default_factory=lambda _: MaxPooling())
@layer_support()
Q
qijun 已提交
1245 1246 1247 1248
def pooling_layer(input,
                  pooling_type=None,
                  name=None,
                  bias_attr=None,
L
Luo Tao 已提交
1249
                  agg_level=AggregateLevel.TO_NO_SEQUENCE,
1250
                  stride=-1,
Z
zhangjinchao01 已提交
1251 1252 1253 1254
                  layer_attr=None):
    """
    Pooling layer for sequence inputs, not used for Image.

1255 1256
    If stride > 0, this layer slides a window whose size is determined by stride,
    and return the pooling value of the window as the output. Thus, a long sequence
X
xuwei06 已提交
1257 1258 1259
    will be shorten.

    The parameter stride specifies the intervals at which to apply the pooling
L
Luo Tao 已提交
1260
    operation. Note that for sequence with sub-sequence, the default value
1261 1262
    of stride is -1.

Z
zhangjinchao01 已提交
1263 1264 1265 1266 1267 1268
    The example usage is:

    .. code-block:: python

       seq_pool = pooling_layer(input=layer,
                                pooling_type=AvgPooling(),
L
Luo Tao 已提交
1269
                                agg_level=AggregateLevel.TO_NO_SEQUENCE)
Z
zhangjinchao01 已提交
1270

L
Luo Tao 已提交
1271 1272
    :param agg_level: AggregateLevel.TO_NO_SEQUENCE or
                      AggregateLevel.TO_SEQUENCE
Z
zhangjinchao01 已提交
1273 1274 1275 1276 1277 1278 1279 1280
    :type agg_level: AggregateLevel
    :param name: layer name.
    :type name: basestring
    :param input: input layer name.
    :type input: LayerOutput
    :param pooling_type: Type of pooling, MaxPooling(default), AvgPooling,
                         SumPooling, SquareRootNPooling.
    :type pooling_type: BasePoolingType|None
L
Luo Tao 已提交
1281
    :param stride: The step size between successive pooling regions.
1282
    :type stride: Int
Z
zhangjinchao01 已提交
1283 1284 1285 1286
    :param bias_attr: Bias parameter attribute. False if no bias.
    :type bias_attr: ParameterAttribute|None|False
    :param layer_attr: The Extra Attributes for layer, such as dropout.
    :type layer_attr: ExtraLayerAttribute|None
D
dangqingqing 已提交
1287
    :return: LayerOutput object.
Y
Yu Yang 已提交
1288
    :rtype: LayerOutput
Z
zhangjinchao01 已提交
1289 1290
    """
    extra_dict = dict()
1291
    # noinspection PyUnresolvedReferences
Z
zhangjinchao01 已提交
1292 1293
    if isinstance(pooling_type, AvgPooling):
        extra_dict['average_strategy'] = pooling_type.strategy
1294 1295 1296 1297
    elif isinstance(pooling_type, MaxPooling) and \
                    pooling_type.output_max_index is not None:
        assert isinstance(pooling_type.output_max_index, bool)
        extra_dict['output_max_index'] = pooling_type.output_max_index
Z
zhangjinchao01 已提交
1298 1299
    extra_dict.update(ExtraLayerAttribute.to_kwargs(layer_attr))

1300 1301 1302
    if agg_level == AggregateLevel.TO_SEQUENCE:
        assert stride == -1

Z
zhangjinchao01 已提交
1303 1304 1305 1306 1307 1308
    Layer(
        name=name,
        type=pooling_type.name,
        inputs=[Input(input.name)],
        bias=ParamAttr.to_bias(bias_attr),
        trans_type=agg_level,
1309
        stride=stride,
Q
qijun 已提交
1310
        **extra_dict)
Z
zhangjinchao01 已提交
1311

Q
qijun 已提交
1312 1313
    return LayerOutput(
        name, pooling_type.name, parents=[input], size=input.size)
Z
zhangjinchao01 已提交
1314

Q
qijun 已提交
1315

Z
zhangjinchao01 已提交
1316 1317
@wrap_bias_attr_default()
@wrap_param_attr_default()
Q
qijun 已提交
1318
@wrap_act_default(param_names=['gate_act'], act=SigmoidActivation())
Z
zhangjinchao01 已提交
1319 1320 1321
@wrap_act_default(param_names=["act", 'state_act'], act=TanhActivation())
@wrap_name_default("lstmemory")
@layer_support(DROPOUT)
Q
qijun 已提交
1322 1323
def lstmemory(input,
              name=None,
1324
              size=None,
Q
qijun 已提交
1325 1326 1327 1328 1329 1330
              reverse=False,
              act=None,
              gate_act=None,
              state_act=None,
              bias_attr=None,
              param_attr=None,
Z
zhangjinchao01 已提交
1331 1332 1333 1334 1335 1336 1337 1338
              layer_attr=None):
    """
    Long Short-term Memory Cell.

    The memory cell was implemented as follow equations.

    ..  math::

L
luotao02 已提交
1339
        i_t & = \\sigma(W_{xi}x_{t} + W_{hi}h_{t-1} + W_{ci}c_{t-1} + b_i)
Z
zhangjinchao01 已提交
1340

L
luotao02 已提交
1341
        f_t & = \\sigma(W_{xf}x_{t} + W_{hf}h_{t-1} + W_{cf}c_{t-1} + b_f)
Z
zhangjinchao01 已提交
1342

L
luotao02 已提交
1343
        c_t & = f_tc_{t-1} + i_t tanh (W_{xc}x_t+W_{hc}h_{t-1} + b_c)
Z
zhangjinchao01 已提交
1344

L
luotao02 已提交
1345
        o_t & = \\sigma(W_{xo}x_{t} + W_{ho}h_{t-1} + W_{co}c_t + b_o)
Z
zhangjinchao01 已提交
1346

L
luotao02 已提交
1347
        h_t & = o_t tanh(c_t)
Z
zhangjinchao01 已提交
1348 1349


C
caoying03 已提交
1350
    NOTE: In PaddlePaddle's implementation, the multiplications
Z
zhangjinchao01 已提交
1351
    :math:`W_{xi}x_{t}` , :math:`W_{xf}x_{t}`,
C
caoying03 已提交
1352 1353 1354 1355
    :math:`W_{xc}x_t`, :math:`W_{xo}x_{t}` are not done in the lstmemory layer,
    so an additional mixed_layer with full_matrix_projection or a fc_layer must
    be included in the configuration file to complete the input-to-hidden
    mappings before lstmemory is called.
Z
zhangjinchao01 已提交
1356

C
caoying03 已提交
1357
    NOTE: This is a low level user interface. You can use network.simple_lstm
Z
zhangjinchao01 已提交
1358 1359
    to config a simple plain lstm layer.

C
caoying03 已提交
1360 1361 1362 1363
    Please refer to **Generating Sequences With Recurrent Neural Networks** for
    more details about LSTM.

    Link_ goes as below.
Z
zhangjinchao01 已提交
1364 1365 1366 1367 1368

    .. _Link: http://arxiv.org/abs/1308.0850

    :param name: The lstmemory layer name.
    :type name: basestring
1369 1370
    :param size: DEPRECATED. size of the lstm cell
    :type size: int
Z
zhangjinchao01 已提交
1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388
    :param input: input layer name.
    :type input: LayerOutput
    :param reverse: is sequence process reversed or not.
    :type reverse: bool
    :param act: activation type, TanhActivation by default. :math:`h_t`
    :type act: BaseActivation
    :param gate_act: gate activation type, SigmoidActivation by default.
    :type gate_act: BaseActivation
    :param state_act: state activation type, TanhActivation by default.
    :type state_act: BaseActivation

    :param bias_attr: Bias attribute. None means default bias. False means no
                      bias.
    :type bias_attr: ParameterAttribute|None|False
    :param param_attr: Parameter Attribute.
    :type param_attr: ParameterAttribute|None|False
    :param layer_attr: Extra Layer attribute
    :type layer_attr: ExtraLayerAttribute|None
D
dangqingqing 已提交
1389
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
1390 1391 1392 1393 1394 1395
    :rtype: LayerOutput
    """

    assert gate_act.support_hppl
    assert state_act.support_hppl
    assert act.support_hppl
1396
    assert input.size is not None and input.size % 4 == 0
1397

1398 1399 1400 1401 1402
    if size is not None:
        if input.size / 4 == size:
            plog = logger.warning
        else:
            plog = logger.fatal
1403 1404 1405
        plog("size of lstmemory layer: %s is automatically set to "
             "size of input layer / 4. The parameter size passing to "
             "this layer is ignored." % (name))
Z
zhangjinchao01 已提交
1406

Q
qijun 已提交
1407 1408 1409 1410 1411 1412 1413 1414 1415 1416
    Layer(
        name=name,
        type=LayerType.LSTMEMORY,
        active_type=act.name,
        active_state_type=state_act.name,
        active_gate_type=gate_act.name,
        reversed=reverse,
        bias=ParamAttr.to_bias(bias_attr),
        inputs=[Input(input.name, **param_attr.attr)],
        **ExtraLayerAttribute.to_kwargs(layer_attr))
Z
zhangjinchao01 已提交
1417

Q
qijun 已提交
1418 1419 1420 1421 1422
    return LayerOutput(
        name,
        LayerType.LSTMEMORY, [input],
        size=input.size / 4,
        reverse=reverse)
1423

Z
zhangjinchao01 已提交
1424 1425 1426

@wrap_bias_attr_default()
@wrap_param_attr_default()
Q
qijun 已提交
1427
@wrap_act_default(param_names=['gate_act'], act=SigmoidActivation())
Z
zhangjinchao01 已提交
1428 1429 1430
@wrap_act_default(param_names=["act"], act=TanhActivation())
@wrap_name_default("gru")
@layer_support(DROPOUT)
Q
qijun 已提交
1431
def grumemory(input,
1432
              size=None,
Q
qijun 已提交
1433 1434 1435 1436 1437 1438
              name=None,
              reverse=False,
              act=None,
              gate_act=None,
              bias_attr=None,
              param_attr=None,
Z
zhangjinchao01 已提交
1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459
              layer_attr=None):
    """
    Gate Recurrent Unit Layer.

    The memory cell was implemented as follow equations.

    1. update gate :math:`z`: defines how much of the previous memory to
    keep around or the unit updates its activations. The update gate
    is computed by:

    ..  math::

        z_t = \\sigma(W_{z}x_{t} + U_{z}h_{t-1} + b_z)

    2. reset gate :math:`r`: determines how to combine the new input with the
    previous memory. The reset gate is computed similarly to the update gate:

    ..  math::

        r_t = \\sigma(W_{r}x_{t} + U_{r}h_{t-1} + b_r)

C
caoying03 已提交
1460 1461
    3. The candidate activation :math:`\\tilde{h_t}` is computed similarly to
    that of the traditional recurrent unit:
Z
zhangjinchao01 已提交
1462 1463 1464 1465 1466

    ..  math::

        {\\tilde{h_t}} = tanh(W x_{t} + U (r_{t} \odot h_{t-1}) + b)

C
caoying03 已提交
1467 1468 1469
    4. The hidden activation :math:`h_t` of the GRU at time t is a linear
    interpolation between the previous activation :math:`h_{t-1}` and the
    candidate activation :math:`\\tilde{h_t}`:
Z
zhangjinchao01 已提交
1470 1471 1472 1473 1474

    ..  math::

        h_t = (1 - z_t) h_{t-1} + z_t {\\tilde{h_t}}

C
caoying03 已提交
1475
    NOTE: In PaddlePaddle's implementation, the multiplication operations
Z
zhangjinchao01 已提交
1476
    :math:`W_{r}x_{t}`, :math:`W_{z}x_{t}` and :math:`W x_t` are not computed in
C
caoying03 已提交
1477 1478 1479
    gate_recurrent layer. Consequently, an additional mixed_layer with
    full_matrix_projection or a fc_layer must be included before grumemory
    is called.
Z
zhangjinchao01 已提交
1480

C
caoying03 已提交
1481 1482 1483
    More details can be found by referring to `Empirical Evaluation of Gated
    Recurrent Neural Networks on Sequence Modeling.
    <https://arxiv.org/abs/1412.3555>`_
Z
zhangjinchao01 已提交
1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494

    The simple usage is:

    .. code-block:: python

       gru = grumemory(input)

    :param name: The gru layer name.
    :type name: None|basestring
    :param input: input layer.
    :type input: LayerOutput.
1495 1496
    :param size: DEPRECATED. size of the gru cell
    :type size: int
1497
    :param reverse: Whether sequence process is reversed or not.
Z
zhangjinchao01 已提交
1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512
    :type reverse: bool
    :param act: activation type, TanhActivation by default. This activation
                affects the :math:`{\\tilde{h_t}}`.
    :type act: BaseActivation
    :param gate_act: gate activation type, SigmoidActivation by default.
                     This activation affects the :math:`z_t` and :math:`r_t`. It is the
                     :math:`\\sigma` in the above formula.
    :type gate_act: BaseActivation
    :param bias_attr: Bias attribute. None means default bias. False means no
                      bias.
    :type bias_attr: ParameterAttribute|None|False
    :param param_attr: Parameter Attribute.
    :type param_attr: ParameterAttribute|None|False
    :param layer_attr: Extra Layer attribute
    :type layer_attr: ExtraLayerAttribute|None
D
dangqingqing 已提交
1513
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
1514 1515 1516 1517
    :rtype: LayerOutput
    """
    assert act.support_hppl
    assert gate_act.support_hppl
1518 1519 1520 1521 1522 1523
    assert input.size is not None and input.size % 3 == 0
    if size is not None:
        if input.size / 3 == size:
            plog = logger.warning
        else:
            plog = logger.fatal
1524 1525 1526
        plog("size of grumemory layer: %s is automatically set to "
             "size of input layer / 3. The parameter size passing to this "
             "layer is ignored." % (name))
Z
zhangjinchao01 已提交
1527

Q
qijun 已提交
1528 1529 1530 1531 1532 1533 1534 1535 1536
    Layer(
        name=name,
        type=LayerType.GRUMEMORY,
        active_type=act.name,
        active_gate_type=gate_act.name,
        reversed=reverse,
        bias=ParamAttr.to_bias(bias_attr),
        inputs=[Input(input.name, **param_attr.attr)],
        **ExtraLayerAttribute.to_kwargs(layer_attr))
Z
zhangjinchao01 已提交
1537

Q
qijun 已提交
1538 1539 1540 1541 1542
    return LayerOutput(
        name,
        LayerType.GRUMEMORY, [input],
        size=input.size / 3,
        reverse=reverse)
1543

Z
zhangjinchao01 已提交
1544 1545 1546

@wrap_name_default()
@layer_support()
Q
qijun 已提交
1547 1548
def last_seq(input,
             name=None,
L
Luo Tao 已提交
1549
             agg_level=AggregateLevel.TO_NO_SEQUENCE,
1550
             stride=-1,
Z
zhangjinchao01 已提交
1551 1552 1553 1554
             layer_attr=None):
    """
    Get Last Timestamp Activation of a sequence.

1555 1556 1557
    If stride > 0, this layer slides a window whose size is determined by stride,
    and return the last value of the window as the output. Thus, a long sequence
    will be shorten. Note that for sequence with sub-sequence, the default value
L
Luo Tao 已提交
1558
    of stride is -1.
1559

L
Luo Tao 已提交
1560 1561 1562 1563 1564 1565
    The simple usage is:

    .. code-block:: python

       seq = last_seq(input=layer)

Z
zhangjinchao01 已提交
1566 1567 1568 1569 1570
    :param agg_level: Aggregated level
    :param name: Layer name.
    :type name: basestring
    :param input: Input layer name.
    :type input: LayerOutput
L
Luo Tao 已提交
1571
    :param stride: The step size between successive pooling regions.
1572
    :type stride: Int
Z
zhangjinchao01 已提交
1573 1574
    :param layer_attr: extra layer attributes.
    :type layer_attr: ExtraLayerAttribute.
D
dangqingqing 已提交
1575
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
1576 1577
    :rtype: LayerOutput
    """
1578 1579 1580 1581 1582 1583
    if input.reverse is not None and input.reverse:
        logger.warning("You are getting the last instance of a sequence that"
                       " is a output of a REVERSED layer. There is no time"
                       " series information at all. Maybe you want to use"
                       " first_seq instead.")

L
Luo Tao 已提交
1584
    if agg_level == AggregateLevel.TO_SEQUENCE:
1585 1586
        assert stride == -1

Z
zhangjinchao01 已提交
1587 1588 1589 1590 1591
    Layer(
        name=name,
        type=LayerType.SEQUENCE_LAST_INSTANCE,
        inputs=[input.name],
        trans_type=agg_level,
1592
        stride=stride,
Q
qijun 已提交
1593 1594 1595 1596 1597 1598
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name,
        LayerType.SEQUENCE_LAST_INSTANCE,
        parents=[input],
        size=input.size)
Z
zhangjinchao01 已提交
1599 1600 1601 1602


@wrap_name_default()
@layer_support()
Q
qijun 已提交
1603 1604
def first_seq(input,
              name=None,
L
Luo Tao 已提交
1605
              agg_level=AggregateLevel.TO_NO_SEQUENCE,
1606
              stride=-1,
Z
zhangjinchao01 已提交
1607 1608 1609 1610
              layer_attr=None):
    """
    Get First Timestamp Activation of a sequence.

1611 1612 1613
    If stride > 0, this layer slides a window whose size is determined by stride,
    and return the first value of the window as the output. Thus, a long sequence
    will be shorten. Note that for sequence with sub-sequence, the default value
L
Luo Tao 已提交
1614
    of stride is -1.
1615

L
Luo Tao 已提交
1616 1617 1618 1619 1620 1621
    The simple usage is:

    .. code-block:: python

       seq = first_seq(input=layer)

Z
zhangjinchao01 已提交
1622 1623 1624 1625 1626
    :param agg_level: aggregation level
    :param name: Layer name.
    :type name: basestring
    :param input: Input layer name.
    :type input: LayerOutput
L
Luo Tao 已提交
1627
    :param stride: The step size between successive pooling regions.
1628
    :type stride: Int
Z
zhangjinchao01 已提交
1629 1630
    :param layer_attr: extra layer attributes.
    :type layer_attr: ExtraLayerAttribute.
D
dangqingqing 已提交
1631
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
1632 1633
    :rtype: LayerOutput
    """
1634 1635 1636 1637 1638 1639 1640

    if input.reverse is not None and not input.reverse:
        logger.warning('You are getting the first instance for a time series,'
                       ' and it is a normal recurrent layer output. There is no'
                       ' time series information at all. Maybe you want to use'
                       ' last_seq instead.')

L
Luo Tao 已提交
1641
    if agg_level == AggregateLevel.TO_SEQUENCE:
1642 1643
        assert stride == -1

Z
zhangjinchao01 已提交
1644 1645 1646 1647 1648
    Layer(
        name=name,
        type=LayerType.SEQUENCE_FIRST_INSTANCE,
        inputs=[input.name],
        trans_type=agg_level,
1649
        stride=stride,
Q
qijun 已提交
1650 1651 1652 1653 1654 1655
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name,
        LayerType.SEQUENCE_FIRST_INSTANCE,
        parents=[input],
        size=input.size)
Z
zhangjinchao01 已提交
1656 1657 1658


class ExpandLevel(object):
1659 1660 1661 1662 1663
    """
    Please refer to AggregateLevel first.

    ExpandLevel supports two modes:

L
Luo Tao 已提交
1664 1665
    - :code:`ExpandLevel.FROM_NO_SEQUENCE` means the expansion acts on
      :code:`NO_SEQUENCE`, which will be expanded to
1666 1667
      :code:`SEQUENCE` or :code:`SUB_SEQUENCE`.

L
Luo Tao 已提交
1668 1669
    - :code:`ExpandLevel.FROM_SEQUENCE` means the expansion acts on
      :code:`SEQUENCE`, which will be expanded to
1670 1671
      :code:`SUB_SEQUENCE`.
    """
L
Luo Tao 已提交
1672 1673
    FROM_NO_SEQUENCE = AggregateLevel.TO_NO_SEQUENCE
    FROM_SEQUENCE = AggregateLevel.TO_SEQUENCE
1674 1675
    # compatible with previous configuration
    FROM_TIMESTEP = FROM_NO_SEQUENCE
Z
zhangjinchao01 已提交
1676

1677

Z
zhangjinchao01 已提交
1678 1679
@wrap_name_default()
@layer_support()
Q
qijun 已提交
1680 1681
def expand_layer(input,
                 expand_as,
Z
zhangjinchao01 已提交
1682 1683
                 name=None,
                 bias_attr=False,
L
Luo Tao 已提交
1684
                 expand_level=ExpandLevel.FROM_NO_SEQUENCE,
Z
zhangjinchao01 已提交
1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695
                 layer_attr=None):
    """
    A layer for "Expand Dense data or (sequence data where the length of each
    sequence is one) to sequence data."

    The example usage is:

    .. code-block:: python

       expand = expand_layer(input=layer1,
                             expand_as=layer2,
L
Luo Tao 已提交
1696
                             expand_level=ExpandLevel.FROM_NO_SEQUENCE)
Z
zhangjinchao01 已提交
1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710

    :param input: Input layer
    :type input: LayerOutput
    :param expand_as: Expand as this layer's sequence info.
    :type expand_as: LayerOutput
    :param name: Layer name.
    :type name: basestring
    :param bias_attr: Bias attribute. None means default bias. False means no
                      bias.
    :type bias_attr: ParameterAttribute|None|False
    :param expand_level: whether input layer is timestep(default) or sequence.
    :type expand_level: ExpandLevel
    :param layer_attr: extra layer attributes.
    :type layer_attr: ExtraLayerAttribute.
D
dangqingqing 已提交
1711
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
1712 1713 1714 1715 1716 1717 1718 1719 1720
    :rtype: LayerOutput
    """

    Layer(
        inputs=[input.name, expand_as.name],
        name=name,
        bias=ParamAttr.to_bias(bias_attr=bias_attr),
        type=LayerType.EXPAND_LAYER,
        trans_type=expand_level,
Q
qijun 已提交
1721 1722 1723 1724 1725 1726
        **ExtraAttr.to_kwargs(layer_attr))
    return LayerOutput(
        name=name,
        size=input.size,
        layer_type=LayerType.EXPAND_LAYER,
        parents=[input, expand_as])
Z
zhangjinchao01 已提交
1727 1728


X
xuwei06 已提交
1729
@wrap_name_default()
X
xuwei06 已提交
1730
@wrap_act_default(act=IdentityActivation())
X
xuwei06 已提交
1731
@layer_support()
X
xuwei06 已提交
1732 1733 1734
def repeat_layer(input,
                 num_repeats,
                 as_row_vector=True,
X
xuwei06 已提交
1735
                 act=None,
X
xuwei06 已提交
1736 1737
                 name=None,
                 layer_attr=None):
X
xuwei06 已提交
1738
    """
X
xuwei06 已提交
1739
    A layer for repeating the input for num_repeats times.
X
xuwei06 已提交
1740

X
xuwei06 已提交
1741
    If as_row_vector:
X
xuwei06 已提交
1742
    .. math::
X
xuwei06 已提交
1743 1744 1745 1746 1747
       y  = [x_1,\cdots, x_n, \cdots, x_1, \cdots, x_n]
    If not as_row_vector:
    .. math::
       y  = [x_1,\cdots, x_1, \cdots, x_n, \cdots, x_n]

X
xuwei06 已提交
1748 1749 1750 1751 1752

    The example usage is:

    .. code-block:: python

L
Luo Tao 已提交
1753
       expand = repeat_layer(input=layer, num_repeats=4)
X
xuwei06 已提交
1754 1755 1756 1757 1758 1759

    :param input: Input layer
    :type input: LayerOutput
    :param num_repeats: Repeat the input so many times
    :type num_repeats: int
    :param name: Layer name.
X
xuwei06 已提交
1760 1761 1762 1763 1764 1765
    :param as_row_vector: True for treating input as row vector and repeating
                          in the column direction.  This is equivalent to apply
                          concat_layer() with num_repeats same input.
                          False for treating input as column vector and repeating
                          in the row direction.
    :type as_row_vector: bool
X
xuwei06 已提交
1766 1767
    :param act: Activation type.
    :type act: BaseActivation
X
xuwei06 已提交
1768 1769 1770 1771 1772 1773 1774 1775 1776 1777
    :type name: basestring
    :param layer_attr: extra layer attributes.
    :type layer_attr: ExtraLayerAttribute.
    :return: LayerOutput object.
    :rtype: LayerOutput
    """

    l = Layer(
        inputs=[input.name],
        name=name,
X
xuwei06 已提交
1778
        active_type=act.name,
X
xuwei06 已提交
1779
        num_filters=num_repeats,
X
xuwei06 已提交
1780
        as_row_vector=as_row_vector,
X
xuwei06 已提交
1781
        type=LayerType.FEATURE_MAP_EXPAND_LAYER,
Q
qijun 已提交
1782 1783 1784 1785 1786
        **ExtraAttr.to_kwargs(layer_attr))
    return LayerOutput(
        name=name,
        size=l.config.size,
        layer_type=LayerType.FEATURE_MAP_EXPAND_LAYER,
X
xuwei06 已提交
1787
        activation=act,
Q
qijun 已提交
1788 1789
        parents=[input])

X
xuwei06 已提交
1790

1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802
@wrap_name_default("seqreshape")
@wrap_act_default(act=IdentityActivation())
@wrap_bias_attr_default(has_bias=False)
@layer_support()
def seq_reshape_layer(input,
                      reshape_size,
                      act=None,
                      name=None,
                      layer_attr=None,
                      bias_attr=None):
    """
    A layer for reshaping the sequence. Assume the input sequence has T instances,
1803
    the dimension of each instance is M, and the input reshape_size is N, then the
1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845
    output sequence has T*M/N instances, the dimension of each instance is N.

    Note that T*M/N must be an integer.

    The example usage is:

    .. code-block:: python

       reshape = seq_reshape_layer(input=layer, reshape_size=4)

    :param input: Input layer.
    :type input: LayerOutput
    :param reshape_size: the size of reshaped sequence.
    :type reshape_size: int
    :param name: Layer name.
    :type name: basestring
    :param act: Activation type.
    :type act: BaseActivation
    :param layer_attr: extra layer attributes.
    :type layer_attr: ExtraLayerAttribute.
    :param bias_attr: The Bias Attribute. If no bias, then pass False or
                      something not type of ParameterAttribute. None will get a
                      default Bias.
    :type bias_attr: ParameterAttribute or None or bool
    :return: LayerOutput object.
    :rtype: LayerOutput
    """

    Layer(
        inputs=[input.name],
        name=name,
        size=reshape_size,
        type=LayerType.SEQUENCE_RESHAPE,
        bias=ParamAttr.to_bias(bias_attr),
        **ExtraAttr.to_kwargs(layer_attr))
    return LayerOutput(
        name=name,
        size=reshape_size,
        layer_type=LayerType.SEQUENCE_RESHAPE,
        parents=[input])


Z
zhangjinchao01 已提交
1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873
@wrap_name_default()
@layer_support()
def interpolation_layer(input, weight, name=None, layer_attr=None):
    """
    This layer is for linear interpolation with two inputs,
    which is used in NEURAL TURING MACHINE.

    .. math::
       y.row[i] = w[i] * x_1.row[i] + (1 - w[i]) * x_2.row[i]

    where :math:`x_1` and :math:`x_2` are two (batchSize x dataDim) inputs,
    :math:`w` is (batchSize x 1) weight vector, and :math:`y` is
    (batchSize x dataDim) output.

    The example usage is:

    .. code-block:: python

       interpolation = interpolation_layer(input=[layer1, layer2], weight=layer3)

    :param input: Input layer.
    :type input: list|tuple
    :param weight: Weight layer.
    :type weight: LayerOutput
    :param name: Layer name.
    :type name: basestring
    :param layer_attr: extra layer attributes.
    :type layer_attr: ExtraLayerAttribute.
D
dangqingqing 已提交
1874
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
1875 1876
    :rtype: LayerOutput
    """
1877
    assert isinstance(input, collections.Sequence)
Z
zhangjinchao01 已提交
1878
    assert len(input) == 2
1879 1880 1881 1882 1883 1884 1885
    assert isinstance(input[0], LayerOutput) and isinstance(input[1],
                                                            LayerOutput)
    if input[0].size is not None and input[1].size is not None:
        assert input[0].size == input[1].size
    assert isinstance(weight, LayerOutput)
    if weight.size is not None:
        assert weight.size == 1
Z
zhangjinchao01 已提交
1886 1887 1888 1889
    Layer(
        name=name,
        type=LayerType.INTERPOLATION_LAYER,
        inputs=[weight.name, input[0].name, input[1].name],
Q
qijun 已提交
1890 1891 1892 1893 1894 1895
        **ExtraAttr.to_kwargs(layer_attr))
    return LayerOutput(
        name,
        LayerType.INTERPOLATION_LAYER,
        parents=[weight, input[0], input[1]],
        size=input[0].size)
Z
zhangjinchao01 已提交
1896 1897


L
liaogang 已提交
1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913
@wrap_name_default()
@layer_support()
def bilinear_interp_layer(input,
                          out_size_x=None,
                          out_size_y=None,
                          name=None,
                          layer_attr=None):
    """
    This layer is to implement bilinear interpolation on conv layer output.

    Please refer to Wikipedia: https://en.wikipedia.org/wiki/Bilinear_interpolation

    The simple usage is:

    .. code-block:: python

L
liaogang 已提交
1914
       bilinear = bilinear_interp_layer(input=layer1, out_size_x=64, out_size_y=64)
X
xuwei06 已提交
1915

L
liaogang 已提交
1916
    :param   input:        A input layer.
L
liaogang 已提交
1917
    :type    input:        LayerOutput.
L
liaogang 已提交
1918
    :param   out_size_x:   bilinear interpolation output width.
X
xuwei06 已提交
1919
    :type    out_size_x:   int|None
L
liaogang 已提交
1920
    :param   out_size_y:   bilinear interpolation output height.
L
liaogang 已提交
1921
    :type    out_size_y:   int|None
L
liaogang 已提交
1922
    :param   name:         The layer's name, which cna not be specified.
L
liaogang 已提交
1923
    :type    name:         None|basestring
L
liaogang 已提交
1924
    :param   layer_attr:   Extra Layer attribute.
L
liaogang 已提交
1925 1926 1927 1928 1929 1930 1931
    :type    layer_attr:   ExtraLayerAttribute
    :return: LayerOutput object.
    :rtype:  LayerOutput
    """
    assert input.layer_type == LayerType.CONV_LAYER
    assert isinstance(input.activation, LinearActivation)
    assert out_size_x > 0 and out_size_y > 0
L
liaogang 已提交
1932
    assert input.num_filters is not None
L
liaogang 已提交
1933
    num_channels = input.num_filters
Q
qijun 已提交
1934 1935 1936 1937 1938 1939 1940
    l = Layer(
        name=name,
        inputs=Input(
            input.name,
            bilinear_interp=BilinearInterp(
                out_size_x=out_size_x,
                out_size_y=out_size_y,
L
Luo Tao 已提交
1941
                channels=num_channels)),
Q
qijun 已提交
1942 1943 1944 1945 1946 1947 1948 1949 1950
        type=LayerType.BILINEAR_INTERP_LAYER,
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name,
        LayerType.BILINEAR_INTERP_LAYER,
        parents=[input],
        num_filters=num_channels,
        size=l.config.size)

L
liaogang 已提交
1951

Z
zhangjinchao01 已提交
1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978
@wrap_name_default()
@layer_support()
def power_layer(input, weight, name=None, layer_attr=None):
    """
    This layer applies a power function to a vector element-wise,
    which is used in NEURAL TURING MACHINE.

    .. math::
       y = x^w

    where :math:`x` is a input vector, :math:`w` is scalar weight,
    and :math:`y` is a output vector.

    The example usage is:

    .. code-block:: python

       power = power_layer(input=layer1, weight=layer2)

    :param input: Input layer.
    :type input: LayerOutput
    :param weight: Weight layer.
    :type weight: LayerOutput
    :param name: Layer name.
    :type name: basestring
    :param layer_attr: extra layer attributes.
    :type layer_attr: ExtraLayerAttribute.
D
dangqingqing 已提交
1979
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
1980 1981
    :rtype: LayerOutput
    """
1982 1983 1984
    assert isinstance(input, LayerOutput) and isinstance(weight, LayerOutput)
    if weight.size is not None:
        assert weight.size == 1
Z
zhangjinchao01 已提交
1985 1986 1987
    Layer(
        name=name,
        type=LayerType.POWER_LAYER,
1988
        inputs=[weight.name, input.name],
Q
qijun 已提交
1989 1990 1991
        **ExtraAttr.to_kwargs(layer_attr))
    return LayerOutput(
        name, LayerType.POWER_LAYER, parents=[input, weight], size=input.size)
Z
zhangjinchao01 已提交
1992 1993 1994 1995 1996 1997


@wrap_name_default()
@layer_support()
def scaling_layer(input, weight, name=None, layer_attr=None):
    """
1998
    A layer for multiplying input vector by weight scalar.
Z
zhangjinchao01 已提交
1999 2000

    .. math::
2001
       y  = w x
Z
zhangjinchao01 已提交
2002

2003 2004 2005 2006 2007
    where :math:`x` is size=dataDim input, :math:`w` is size=1 weight,
    and :math:`y` is size=dataDim output.

    Note that the above computation is for one sample. Multiple samples are
    processed in one batch.
Z
zhangjinchao01 已提交
2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022

    The example usage is:

    .. code-block:: python

       scale = scaling_layer(input=layer1, weight=layer2)

    :param input: Input layer.
    :type input: LayerOutput
    :param weight: Weight layer.
    :type weight: LayerOutput
    :param name: Layer name.
    :type name: basestring
    :param layer_attr: extra layer attributes.
    :type layer_attr: ExtraLayerAttribute.
D
dangqingqing 已提交
2023
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
2024 2025
    :rtype: LayerOutput
    """
2026 2027 2028
    assert isinstance(weight, LayerOutput) and isinstance(input, LayerOutput)
    if weight.size is not None:
        assert weight.size == 1
Z
zhangjinchao01 已提交
2029 2030 2031 2032
    Layer(
        name=name,
        type=LayerType.SCALING_LAYER,
        inputs=[weight.name, input.name],
Q
qijun 已提交
2033 2034 2035
        **ExtraAttr.to_kwargs(layer_attr))
    return LayerOutput(
        name, LayerType.SCALING_LAYER, parents=[weight, input], size=input.size)
Z
zhangjinchao01 已提交
2036 2037 2038 2039 2040 2041


@wrap_name_default()
@layer_support()
def trans_layer(input, name=None, layer_attr=None):
    """
2042
    A layer for transposing a minibatch matrix.
Z
zhangjinchao01 已提交
2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060

    .. math::
       y = x^\mathrm{T}

    where :math:`x` is (M x N) input, and :math:`y` is (N x M) output.

    The example usage is:

    .. code-block:: python

       trans = trans_layer(input=layer)

    :param input: Input layer.
    :type input: LayerOutput
    :param name: Layer name.
    :type name: basestring
    :param layer_attr: extra layer attributes.
    :type layer_attr: ExtraLayerAttribute.
D
dangqingqing 已提交
2061
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
2062 2063 2064 2065 2066 2067
    :rtype: LayerOutput
    """
    Layer(
        name=name,
        type=LayerType.TRANS_LAYER,
        inputs=[input.name],
Q
qijun 已提交
2068 2069 2070
        **ExtraAttr.to_kwargs(layer_attr))
    return LayerOutput(
        name, LayerType.TRANS_LAYER, parents=[input], size=input.size)
Z
zhangjinchao01 已提交
2071 2072


2073 2074
@wrap_name_default()
@layer_support()
H
Haonan 已提交
2075
def rotate_layer(input, height, width, name=None, layer_attr=None):
2076
    """
H
Haonan 已提交
2077 2078
    A layer for rotating 90 degrees (clock-wise) for each feature channel,
    usually used when the input sample is some image or feature map.
2079 2080

    .. math::
H
Haonan 已提交
2081
       y(j,i,:) = x(M-i-1,j,:)
2082

H
Haonan 已提交
2083
    where :math:`x` is (M x N x C) input, and :math:`y` is (N x M x C) output.
2084 2085 2086 2087 2088 2089

    The example usage is:

    .. code-block:: python

       rot = rotate_layer(input=layer,
H
Haonan 已提交
2090 2091
                          height=100,
                          width=100)
2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104

    :param input: Input layer.
    :type input: LayerOutput
    :param height: The height of the sample matrix
    :type height: int
    :param name: Layer name.
    :type name: basestring
    :param layer_attr: extra layer attributes.
    :type layer_attr: ExtraLayerAttribute.
    :return: LayerOutput object.
    :rtype: LayerOutput
    """
    assert isinstance(input, LayerOutput)
H
Haonan 已提交
2105 2106 2107
    l = Layer(
        name=name,
        height=height,
H
Haonan 已提交
2108
        width=width,
H
Haonan 已提交
2109 2110 2111 2112 2113 2114 2115 2116
        type=LayerType.ROTATE_LAYER,
        inputs=[input.name],
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name=name,
        layer_type=LayerType.ROTATE_LAYER,
        parents=[input],
        size=l.config.size)
2117 2118


Z
zhangjinchao01 已提交
2119 2120
@wrap_name_default()
@layer_support()
2121
def cos_sim(a, b, scale=1, size=1, name=None, layer_attr=None):
Z
zhangjinchao01 已提交
2122 2123 2124 2125
    """
    Cosine Similarity Layer. The cosine similarity equation is here.

    ..  math::
D
dangqingqing 已提交
2126
        similarity = cos(\\theta) = {\\mathbf{a} \\cdot \\mathbf{b}
2127 2128 2129 2130 2131
        \\over \\|\\mathbf{a}\\| \\|\\mathbf{b}\\|}

    The size of a is M, size of b is M*N,
    Similarity will be calculated N times by step M. The output size is
    N. The scale will be multiplied to similarity.
Z
zhangjinchao01 已提交
2132

2133 2134
    Note that the above computation is for one sample. Multiple samples are
    processed in one batch.
Z
zhangjinchao01 已提交
2135

L
Luo Tao 已提交
2136 2137 2138 2139 2140 2141
    The example usage is:

    .. code-block:: python

       cos = cos_sim(a=layer1, b=layer2, size=3)

Z
zhangjinchao01 已提交
2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153
    :param name: layer name
    :type name: basestring
    :param a: input layer a
    :type a: LayerOutput
    :param b: input layer b
    :type b: LayerOutput
    :param scale: scale for cosine value. default is 5.
    :type scale: float
    :param size: layer size. NOTE size_a * size should equal size_b.
    :type size: int
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
D
dangqingqing 已提交
2154
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
2155 2156
    :rtype: LayerOutput
    """
2157
    assert isinstance(a, LayerOutput) and isinstance(b, LayerOutput)
2158 2159 2160 2161 2162 2163
    if size == 1:
        Layer(
            name=name,
            type=LayerType.COSINE_SIM,
            cos_scale=scale,
            inputs=[a.name, b.name],
Q
qijun 已提交
2164
            **ExtraLayerAttribute.to_kwargs(layer_attr))
2165
    else:
2166 2167
        if a.size is not None and b.size is not None:
            assert size == b.size / a.size
2168 2169 2170 2171 2172 2173
        Layer(
            name=name,
            type=LayerType.COSINE_SIM_VEC,
            size=size,
            cos_scale=scale,
            inputs=[a.name, b.name],
Q
qijun 已提交
2174
            **ExtraLayerAttribute.to_kwargs(layer_attr))
X
xuwei06 已提交
2175
    return LayerOutput(name, LayerType.COSINE_SIM, parents=[a, b], size=size)
Z
zhangjinchao01 已提交
2176

2177

Z
zhangjinchao01 已提交
2178 2179
@wrap_name_default()
@wrap_bias_attr_default(has_bias=True)
2180
@wrap_param_attr_default()
Z
zhangjinchao01 已提交
2181
@layer_support()
Q
qijun 已提交
2182 2183
def hsigmoid(input,
             label,
2184
             num_classes=None,
Q
qijun 已提交
2185 2186 2187 2188
             name=None,
             bias_attr=None,
             param_attr=None,
             layer_attr=None):
Z
zhangjinchao01 已提交
2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199
    """
    Organize the classes into a binary tree. At each node, a sigmoid function
    is used to calculate the probability of belonging to the right branch.
    This idea is from "F. Morin, Y. Bengio (AISTATS 05):
    Hierarchical Probabilistic Neural Network Language Model."

    The example usage is:

    ..  code-block:: python

        cost = hsigmoid(input=[layer1, layer2],
2200
                        label=data_layer)
Z
zhangjinchao01 已提交
2201 2202 2203 2204 2205 2206 2207

    :param input: Input layers. It could be a LayerOutput or list/tuple of
                 LayerOutput.
    :type input: LayerOutput|list|tuple
    :param label: Label layer.
    :type label: LayerOutput
    :param num_classes: number of classes.
2208
    :type num_classes: int|None
L
luotao02 已提交
2209 2210
    :param name: layer name
    :type name: basestring
Z
zhangjinchao01 已提交
2211 2212 2213
    :param bias_attr: Bias attribute. None means default bias.
                      False means no bias.
    :type bias_attr: ParameterAttribute|False
2214 2215
    :param param_attr: Parameter Attribute. None means default parameter.
    :type param_attr: ParameterAttribute|None
Z
zhangjinchao01 已提交
2216 2217
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
D
dangqingqing 已提交
2218
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
2219 2220 2221 2222
    :rtype: LayerOutput
    """
    if isinstance(input, LayerOutput):
        input = [input]
2223 2224 2225 2226 2227 2228 2229 2230 2231
        if not isinstance(param_attr, collections.Sequence):
            param_attr = [param_attr]
    else:
        if not isinstance(param_attr, collections.Sequence):
            param_attr = [param_attr] * len(input)
        else:
            assert len(param_attr) == len(input)

    assert isinstance(input, collections.Sequence)
Z
zhangjinchao01 已提交
2232 2233 2234
    assert isinstance(label, LayerOutput)
    assert label.layer_type == LayerType.DATA

2235 2236 2237 2238 2239
    if num_classes is None:
        num_classes = label.size
    if num_classes is None or num_classes <= 2:
        raise ValueError("hsigmoid label size must larger than 2.")

Z
zhangjinchao01 已提交
2240 2241
    ipts_for_layer = []
    parents = []
2242
    for each_input, each_param_attr in zip(input, param_attr):
Z
zhangjinchao01 已提交
2243
        assert isinstance(each_input, LayerOutput)
2244
        ipts_for_layer.append(Input(each_input.name, **each_param_attr.attr))
Z
zhangjinchao01 已提交
2245 2246 2247 2248
        parents.append(each_input)
    ipts_for_layer.append(label.name)
    parents.append(label)

X
xuwei06 已提交
2249
    l = Layer(
Z
zhangjinchao01 已提交
2250 2251 2252 2253 2254
        name=name,
        type=LayerType.HSIGMOID,
        num_classes=num_classes,
        bias=ParamAttr.to_bias(bias_attr),
        inputs=ipts_for_layer,
Q
qijun 已提交
2255 2256 2257
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name, LayerType.HSIGMOID, parents=parents, size=l.config.size)
Z
zhangjinchao01 已提交
2258

2259

Z
zhangjinchao01 已提交
2260 2261 2262 2263 2264
@wrap_name_default("conv")
@wrap_param_attr_default()
@wrap_bias_attr_default()
@wrap_act_default(act=ReluActivation())
@layer_support(DROPOUT)
Q
qijun 已提交
2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280
def img_conv_layer(input,
                   filter_size,
                   num_filters,
                   name=None,
                   num_channels=None,
                   act=None,
                   groups=1,
                   stride=1,
                   padding=0,
                   bias_attr=None,
                   param_attr=None,
                   shared_biases=True,
                   layer_attr=None,
                   filter_size_y=None,
                   stride_y=None,
                   padding_y=None,
2281 2282
                   trans=False,
                   layer_type=None):
Z
zhangjinchao01 已提交
2283
    """
2284
    Convolution layer for image. Paddle can support both square and non-square
2285
    input currently.
Z
zhangjinchao01 已提交
2286 2287 2288 2289

    The details of convolution layer, please refer UFLDL's `convolution
    <http://ufldl.stanford.edu/tutorial/supervised/
    FeatureExtractionUsingConvolution/>`_ .
X
xuwei06 已提交
2290

2291
    Convolution Transpose (deconv) layer for image. Paddle can support both square
2292
    and non-square input currently.
2293

X
xuwei06 已提交
2294
    The details of convolution transpose layer,
2295 2296 2297
    please refer to the following explanation and references therein
    <http://datascience.stackexchange.com/questions/6107/
    what-are-deconvolutional-layers/>`_ .
Z
zhangjinchao01 已提交
2298 2299 2300 2301
    The num_channel means input image's channel number. It may be 1 or 3 when
    input is raw pixels of image(mono or RGB), or it may be the previous layer's
    num_filters * num_group.

C
caoying03 已提交
2302 2303 2304
    There are several group of filter in PaddlePaddle implementation.
    Each group will process some channel of the inputs. For example, if an input
    num_channel = 256, group = 4, num_filter=32, the PaddlePaddle will create
Z
zhangjinchao01 已提交
2305
    32*4 = 128 filters to process inputs. The channels will be split into 4
C
caoying03 已提交
2306 2307
    pieces. First 256/4 = 64 channels will process by first 32 filters. The
    rest channels will be processed by rest group of filters.
Z
zhangjinchao01 已提交
2308

L
Luo Tao 已提交
2309 2310 2311 2312 2313 2314 2315 2316 2317 2318
    The example usage is:

    ..  code-block:: python

        conv = img_conv_layer(input=data, filter_size=1, filter_size_y=1,
                              num_channels=8,
                              num_filters=16, stride=1,
                              bias_attr=False,
                              act=ReluActivation())

Z
zhangjinchao01 已提交
2319 2320 2321 2322
    :param name: Layer name.
    :type name: basestring
    :param input: Layer Input.
    :type input: LayerOutput
2323 2324 2325
    :param filter_size: The x dimension of a filter kernel. Or input a tuple for
                        two image dimension.
    :type filter_size: int|tuple|list
C
caoying03 已提交
2326 2327 2328
    :param filter_size_y: The y dimension of a filter kernel. Since PaddlePaddle
                        currently supports rectangular filters, the filter's
                        shape will be (filter_size, filter_size_y).
2329
    :type filter_size_y: int|None
Z
zhangjinchao01 已提交
2330 2331 2332 2333 2334
    :param num_filters: Each filter group's number of filter
    :param act: Activation type. Default is tanh
    :type act: BaseActivation
    :param groups: Group size of filters.
    :type groups: int
2335 2336 2337
    :param stride: The x dimension of the stride. Or input a tuple for two image
                   dimension.
    :type stride: int|tuple|list
Z
zhangjinchao01 已提交
2338 2339
    :param stride_y: The y dimension of the stride.
    :type stride_y: int
2340 2341 2342
    :param padding: The x dimension of the padding. Or input a tuple for two
                    image dimension
    :type padding: int|tuple|list
Z
zhangjinchao01 已提交
2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356
    :param padding_y: The y dimension of the padding.
    :type padding_y: int
    :param bias_attr: Convolution bias attribute. None means default bias.
                      False means no bias.
    :type bias_attr: ParameterAttribute|False
    :param num_channels: number of input channels. If None will be set
                        automatically from previous output.
    :type num_channels: int
    :param param_attr: Convolution param attribute. None means default attribute
    :type param_attr: ParameterAttribute
    :param shared_biases: Is biases will be shared between filters or not.
    :type shared_biases: bool
    :param layer_attr: Layer Extra Attribute.
    :type layer_attr: ExtraLayerAttribute
2357 2358
    :param trans: true if it is a convTransLayer, false if it is a convLayer
    :type trans: bool
2359
    :param layer_type: specify the layer_type, default is None. If trans=True,
2360 2361
                       layer_type has to be "exconvt" or "cudnn_convt",
                       otherwise layer_type has to be either "exconv" or
2362
                       "cudnn_conv"
2363
    :type layer_type: String
D
dangqingqing 已提交
2364
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
2365 2366 2367 2368 2369
    :rtype: LayerOutput
    """
    if num_channels is None:
        assert input.num_filters is not None
        num_channels = input.num_filters
2370

Z
zhangjinchao01 已提交
2371
    if filter_size_y is None:
2372 2373 2374 2375 2376 2377
        if isinstance(filter_size, collections.Sequence):
            assert len(filter_size) == 2
            filter_size, filter_size_y = filter_size
        else:
            filter_size_y = filter_size

Z
zhangjinchao01 已提交
2378
    if stride_y is None:
2379 2380 2381 2382 2383 2384
        if isinstance(stride, collections.Sequence):
            assert len(stride) == 2
            stride, stride_y = stride
        else:
            stride_y = stride

Z
zhangjinchao01 已提交
2385
    if padding_y is None:
2386 2387 2388 2389 2390 2391 2392 2393
        if isinstance(padding, collections.Sequence):
            assert len(padding) == 2
            padding, padding_y = padding
        else:
            padding_y = padding

    if param_attr.attr.get('initial_smart'):
        # special initial for conv layers.
Q
qijun 已提交
2394
        init_w = (2.0 / (filter_size**2 * num_channels))**0.5
2395 2396 2397 2398
        param_attr.attr["initial_mean"] = 0.0
        param_attr.attr["initial_std"] = init_w
        param_attr.attr["initial_strategy"] = 0
        param_attr.attr["initial_smart"] = False
2399

2400 2401
    if layer_type:
        if trans:
2402
            assert layer_type in ["exconvt", "cudnn_convt"]
2403 2404 2405 2406 2407
        else:
            assert layer_type in ["exconv", "cudnn_conv"]
        lt = layer_type
    else:
        lt = LayerType.CONVTRANS_LAYER if trans else LayerType.CONV_LAYER
Q
qijun 已提交
2408

X
xuwei06 已提交
2409
    l = Layer(
Z
zhangjinchao01 已提交
2410
        name=name,
Q
qijun 已提交
2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422
        inputs=Input(
            input.name,
            conv=Conv(
                filter_size=filter_size,
                padding=padding,
                stride=stride,
                channels=num_channels,
                groups=groups,
                filter_size_y=filter_size_y,
                padding_y=padding_y,
                stride_y=stride_y),
            **param_attr.attr),
Z
zhangjinchao01 已提交
2423 2424 2425 2426
        active_type=act.name,
        num_filters=num_filters,
        bias=ParamAttr.to_bias(bias_attr),
        shared_biases=shared_biases,
2427
        type=lt,
Q
qijun 已提交
2428 2429 2430 2431 2432 2433 2434 2435
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name,
        lt,
        parents=[input],
        activation=act,
        num_filters=num_filters,
        size=l.config.size)
Z
zhangjinchao01 已提交
2436 2437 2438 2439


@wrap_name_default("pool")
@layer_support()
Q
qijun 已提交
2440 2441 2442 2443 2444 2445 2446 2447 2448 2449
def img_pool_layer(input,
                   pool_size,
                   name=None,
                   num_channels=None,
                   pool_type=None,
                   stride=1,
                   padding=0,
                   layer_attr=None,
                   pool_size_y=None,
                   stride_y=None,
2450 2451
                   padding_y=None,
                   ceil_mode=True):
Z
zhangjinchao01 已提交
2452 2453 2454 2455 2456 2457 2458
    """
    Image pooling Layer.

    The details of pooling layer, please refer ufldl's pooling_ .

    .. _pooling: http://ufldl.stanford.edu/tutorial/supervised/Pooling/

L
Luo Tao 已提交
2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486
    - ceil_mode=True:

    ..  math::

        w = 1 + int(ceil(input\_width + 2 * padding - pool\_size) / float(stride))
        h = 1 + int(ceil(input\_height + 2 * padding\_y - pool\_size\_y) / float(stride\_y))

    - ceil_mode=False:

    ..  math::

        w = 1 + int(floor(input\_width + 2 * padding - pool\_size) / float(stride))
        h = 1 + int(floor(input\_height + 2 * padding\_y - pool\_size\_y) / float(stride\_y))

    The example usage is:

    ..  code-block:: python

        maxpool = img_pool_layer(input=conv,
                                 pool_size=3,
                                 pool_size_y=5,
                                 num_channels=8,
                                 stride=1,
                                 stride_y=2,
                                 padding=1,
                                 padding_y=2,
                                 pool_type=MaxPooling())

2487
    :param padding: pooling padding width.
Z
zhangjinchao01 已提交
2488
    :type padding: int
2489 2490
    :param padding_y: pooling padding height. It's equal to padding by default.
    :type padding_y: int|None
Z
zhangjinchao01 已提交
2491 2492 2493 2494
    :param name: name of pooling layer
    :type name: basestring.
    :param input: layer's input
    :type input: LayerOutput
2495
    :param pool_size: pooling window width
Z
zhangjinchao01 已提交
2496
    :type pool_size: int
2497 2498
    :param pool_size_y: pooling window height. It's eaqual to pool_size by default.
    :type pool_size_y: int|None
Z
zhangjinchao01 已提交
2499 2500
    :param num_channels: number of input channel.
    :type num_channels: int
2501
    :param pool_type: pooling type. MaxPooling or AvgPooling. Default is
Z
zhangjinchao01 已提交
2502 2503
                      MaxPooling.
    :type pool_type: BasePoolingType
2504
    :param stride: stride width of pooling.
Z
zhangjinchao01 已提交
2505
    :type stride: int
2506 2507
    :param stride_y: stride height of pooling. It is equal to stride by default.
    :type stride_y: int|None
Z
zhangjinchao01 已提交
2508 2509
    :param layer_attr: Extra Layer attribute.
    :type layer_attr: ExtraLayerAttribute
2510 2511 2512 2513
    :param ceil_mode: Wether to use ceil mode to calculate output height and with.
                      Defalut is True. If set false, Otherwise use floor.

    :type ceil_mode: bool
D
dangqingqing 已提交
2514 2515
    :return: LayerOutput object.
    :rtype: LayerOutput
Z
zhangjinchao01 已提交
2516 2517 2518 2519 2520 2521 2522 2523 2524 2525
    """
    if num_channels is None:
        assert input.num_filters is not None
        num_channels = input.num_filters

    if pool_type is None:
        pool_type = MaxPooling()
    elif isinstance(pool_type, AvgPooling):
        pool_type.name = 'avg'

2526
    type_name = pool_type.name + '-projection' \
Y
Yu Yang 已提交
2527
        if (
Y
Yu Yang 已提交
2528
        isinstance(pool_type, AvgPooling) or isinstance(pool_type, MaxPooling)) \
Y
Yu Yang 已提交
2529
        else pool_type.name
2530 2531 2532 2533 2534

    pool_size_y = pool_size if pool_size_y is None else pool_size_y
    stride_y = stride if stride_y is None else stride_y
    padding_y = padding if padding_y is None else padding_y

X
xuwei06 已提交
2535
    l = Layer(
Z
zhangjinchao01 已提交
2536 2537
        name=name,
        type=LayerType.POOL_LAYER,
Q
qijun 已提交
2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549
        inputs=[
            Input(
                input.name,
                pool=Pool(
                    pool_type=type_name,
                    channels=num_channels,
                    size_x=pool_size,
                    start=None,
                    stride=stride,
                    padding=padding,
                    size_y=pool_size_y,
                    stride_y=stride_y,
L
Luo Tao 已提交
2550
                    padding_y=padding_y))
Q
qijun 已提交
2551
        ],
2552
        ceil_mode=ceil_mode,
Q
qijun 已提交
2553 2554 2555 2556 2557 2558 2559
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name,
        LayerType.POOL_LAYER,
        parents=[input],
        num_filters=num_channels,
        size=l.config.size)
Z
zhangjinchao01 已提交
2560 2561


Q
qijun 已提交
2562 2563
@wrap_name_default("spp")
@layer_support()
Q
qijun 已提交
2564 2565 2566 2567 2568 2569
def spp_layer(input,
              name=None,
              num_channels=None,
              pool_type=None,
              pyramid_height=None,
              layer_attr=None):
Q
qijun 已提交
2570 2571 2572 2573 2574
    """
    Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition.
    The details please refer to
    `Kaiming He's paper <https://arxiv.org/abs/1406.4729>`_.

L
Luo Tao 已提交
2575 2576 2577 2578
    The example usage is:

    ..  code-block:: python

2579 2580 2581
        spp = spp_layer(input=data,
                        pyramid_height=2,
                        num_channels=16,
L
Luo Tao 已提交
2582 2583
                        pool_type=MaxPooling())

Q
qijun 已提交
2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611
    :param name: layer name.
    :type name: basestring
    :param input: layer's input.
    :type input: LayerOutput
    :param num_channels: number of input channel.
    :type num_channels: int
    :param pool_type: Pooling type. MaxPooling or AveragePooling. Default is MaxPooling.
    :type scale: BasePoolingType
    :param pyramid_height: pyramid height.
    :type pyramid_height: int
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
    :return: LayerOutput object.
    :rtype: LayerOutput
    """
    if num_channels is None:
        assert input.num_filters is not None
        num_channels = input.num_filters

    if pool_type is None:
        pool_type = MaxPooling()
    elif isinstance(pool_type, AvgPooling):
        pool_type.name = 'avg'

    type_name = pool_type.name
    if (isinstance(pool_type, AvgPooling) or isinstance(pool_type, MaxPooling)):
        type_name += '-projection'

Q
qijun 已提交
2612
    l = Layer(
Q
qijun 已提交
2613 2614
        name=name,
        type=LayerType.SPP_LAYER,
Q
qijun 已提交
2615 2616 2617 2618 2619
        inputs=Input(
            input.name,
            spp=SpatialPyramidPool(
                pool_type=type_name,
                channels=num_channels,
L
Luo Tao 已提交
2620
                pyramid_height=pyramid_height)),
Q
qijun 已提交
2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name,
        layer_type=LayerType.SPP_LAYER,
        parents=[input],
        num_filters=num_channels,
        size=l.config.size)


def __img_norm_layer__(name, input, size, norm_type, scale, power, num_channels,
                       blocked, layer_attr):
Z
zhangjinchao01 已提交
2632 2633 2634 2635
    if num_channels is None:
        assert input.num_filters is not None
        num_channels = input.num_filters

X
xuwei06 已提交
2636
    l = Layer(
Q
qijun 已提交
2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655
        name=name,
        type=LayerType.NORM_LAYER,
        inputs=Input(
            input.name,
            norm=Norm(
                norm_type=norm_type,
                channels=num_channels,
                size=size,
                scale=scale,
                pow=power,
                blocked=blocked)),
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name,
        layer_type=LayerType.NORM_LAYER,
        parents=[input],
        num_filters=num_channels,
        img_norm_type=norm_type,
        size=l.config.size)
Z
zhangjinchao01 已提交
2656 2657 2658 2659


@wrap_name_default("crmnorm")
@layer_support()
Q
qijun 已提交
2660 2661 2662 2663 2664 2665
def img_cmrnorm_layer(input,
                      size,
                      scale=0.0128,
                      power=0.75,
                      name=None,
                      num_channels=None,
2666
                      layer_attr=None):
Z
zhangjinchao01 已提交
2667
    """
2668
    Response normalization across feature maps.
D
dangqingqing 已提交
2669 2670
    The details please refer to
    `Alex's paper <http://www.cs.toronto.edu/~fritz/absps/imagenet.pdf>`_.
Z
zhangjinchao01 已提交
2671

L
Luo Tao 已提交
2672 2673 2674
    The example usage is:

    ..  code-block:: python
2675

L
Luo Tao 已提交
2676 2677
        norm = img_cmrnorm_layer(input=net, size=5)

Z
zhangjinchao01 已提交
2678
    :param name: layer name.
D
dangqingqing 已提交
2679
    :type name: None|basestring
Z
zhangjinchao01 已提交
2680 2681
    :param input: layer's input.
    :type input: LayerOutput
2682
    :param size: Normalize in number of :math:`size` feature maps.
Z
zhangjinchao01 已提交
2683
    :type size: int
D
dangqingqing 已提交
2684
    :param scale: The hyper-parameter.
Z
zhangjinchao01 已提交
2685
    :type scale: float
D
dangqingqing 已提交
2686
    :param power: The hyper-parameter.
Z
zhangjinchao01 已提交
2687 2688 2689 2690 2691
    :type power: float
    :param num_channels: input layer's filers number or channels. If
                         num_channels is None, it will be set automatically.
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
D
dangqingqing 已提交
2692
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
2693 2694 2695
    :rtype: LayerOutput
    """
    return __img_norm_layer__(name, input, size, "cmrnorm-projection", scale,
2696
                              power, num_channels, 0, layer_attr)
Z
zhangjinchao01 已提交
2697 2698 2699


@wrap_bias_attr_default()
2700 2701
@wrap_param_attr_default(
    default_factory=lambda _: ParamAttr(initial_mean=1.0, initial_std=0.))
Z
zhangjinchao01 已提交
2702 2703 2704
@wrap_act_default(act=ReluActivation())
@wrap_name_default("batch_norm")
@layer_support(DROPOUT)
Q
qijun 已提交
2705 2706 2707 2708 2709 2710 2711
def batch_norm_layer(input,
                     act=None,
                     name=None,
                     num_channels=None,
                     bias_attr=None,
                     param_attr=None,
                     layer_attr=None,
Z
zhangjinchao01 已提交
2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732
                     batch_norm_type=None,
                     moving_average_fraction=0.9,
                     use_global_stats=None):
    """
    Batch Normalization Layer. The notation of this layer as follow.

    :math:`x` 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

    The details of batch normalization please refer to this
    `paper <http://arxiv.org/abs/1502.03167>`_.

L
Luo Tao 已提交
2733 2734 2735
    The example usage is:

    ..  code-block:: python
2736

L
Luo Tao 已提交
2737 2738
        norm = batch_norm_layer(input=net, act=ReluActivation())

Z
zhangjinchao01 已提交
2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752
    :param name: layer name.
    :type name: basestring
    :param input: batch normalization input. Better be linear activation.
                Because there is an activation inside batch_normalization.
    :type input: LayerOutput
    :param batch_norm_type: We have batch_norm and cudnn_batch_norm. batch_norm
                            supports both CPU and GPU. cudnn_batch_norm requires
                            cuDNN version greater or equal to v4 (>=v4). But
                            cudnn_batch_norm is faster and needs less memory
                            than batch_norm. By default (None), we will
                            automaticly select cudnn_batch_norm for GPU and
                            batch_norm for CPU. Otherwise, select batch norm
                            type based on the specified type. If you use cudnn_batch_norm,
                            we suggested you use latest version, such as v5.1.
2753
    :type batch_norm_type: None|string, None or "batch_norm" or "cudnn_batch_norm"
Z
zhangjinchao01 已提交
2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780
    :param act: Activation Type. Better be relu. Because batch
                     normalization will normalize input near zero.
    :type act: BaseActivation
    :param num_channels: num of image channels or previous layer's number of
                         filters. None will automatically get from layer's
                         input.
    :type num_channels: int
    :param bias_attr: :math:`\\beta`, better be zero when initialize. So the
                      initial_std=0, initial_mean=1 is best practice.
    :type bias_attr: ParameterAttribute
    :param param_attr: :math:`\\gamma`, better be one when initialize. So the
                       initial_std=0, initial_mean=1 is best practice.
    :type param_attr: ParameterAttribute
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
    :param use_global_stats: whether use moving mean/variance statistics
                             during testing peroid. If None or True,
                             it will use moving mean/variance statistics during
                             testing. If False, it will use the mean
                             and variance of current batch of test data for
                             testing.
    :type use_global_stats: bool|None.
    :param moving_average_fraction: Factor used in the moving average
                                   computation, referred to as facotr,
                                   :math:`runningMean = newMean*(1-factor)
                                   + runningMean*factor`
    :type moving_average_fraction: float.
D
dangqingqing 已提交
2781
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800
    :rtype: LayerOutput
    """
    if not isinstance(act, ReluActivation):
        logger.log(logging.WARN,
                   "%s is not recommend for batch normalization's activation, "
                   "maybe the relu is better" % act.name)

    if not isinstance(input.activation, LinearActivation):
        logger.log(logging.WARN,
                   "The activation should be inside batch normalization, the "
                   "previous layer's activation may be Linear")

    if num_channels is None:
        if input.num_filters is not None:
            num_channels = input.num_filters
        else:
            num_channels = input.size
    assert (batch_norm_type is None) or (batch_norm_type == "batch_norm") or \
           (batch_norm_type == "cudnn_batch_norm")
X
xuwei06 已提交
2801
    l = Layer(
Z
zhangjinchao01 已提交
2802
        name=name,
Q
qijun 已提交
2803 2804
        inputs=Input(
            input.name, image=Image(channels=num_channels), **param_attr.attr),
Z
zhangjinchao01 已提交
2805 2806 2807 2808 2809 2810
        active_type=act.name,
        type=LayerType.BATCH_NORM_LAYER,
        batch_norm_type=batch_norm_type,
        bias=ParamAttr.to_bias(bias_attr),
        moving_average_fraction=moving_average_fraction,
        use_global_stats=use_global_stats,
Q
qijun 已提交
2811
        **ExtraLayerAttribute.to_kwargs(layer_attr))
Z
zhangjinchao01 已提交
2812

Q
qijun 已提交
2813 2814 2815 2816 2817 2818 2819
    return LayerOutput(
        name=name,
        layer_type=LayerType.BATCH_NORM_LAYER,
        parents=[input],
        activation=act,
        num_filters=num_channels,
        size=l.config.size)
Z
zhangjinchao01 已提交
2820 2821 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


@wrap_name_default()
@layer_support()
def sum_to_one_norm_layer(input, name=None, layer_attr=None):
    """
    A layer for sum-to-one normalization,
    which is used in NEURAL TURING MACHINE.

    .. math::
       out[i] = \\frac {in[i]} {\sum_{k=1}^N in[k]}

    where :math:`in` is a (batchSize x dataDim) input vector,
    and :math:`out` is a (batchSize x dataDim) output vector.

    The example usage is:

    .. code-block:: python

       sum_to_one_norm = sum_to_one_norm_layer(input=layer)

    :param input: Input layer.
    :type input: LayerOutput
    :param name: Layer name.
    :type name: basestring
    :param layer_attr: extra layer attributes.
    :type layer_attr: ExtraLayerAttribute.
D
dangqingqing 已提交
2847
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
2848 2849 2850 2851 2852 2853
    :rtype: LayerOutput
    """
    Layer(
        name=name,
        type=LayerType.SUM_TO_ONE_NORM_LAYER,
        inputs=[input.name],
Q
qijun 已提交
2854 2855 2856
        **ExtraAttr.to_kwargs(layer_attr))
    return LayerOutput(
        name, LayerType.SUM_TO_ONE_NORM_LAYER, parents=[input], size=input.size)
Z
zhangjinchao01 已提交
2857 2858 2859 2860 2861 2862


@wrap_name_default("addto")
@wrap_act_default(act=LinearActivation())
@wrap_bias_attr_default(has_bias=False)
@layer_support(DROPOUT)
Q
qijun 已提交
2863
def addto_layer(input, act=None, name=None, bias_attr=None, layer_attr=None):
Z
zhangjinchao01 已提交
2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885
    """
    AddtoLayer.

    ..  math::

        y = f(\\sum_{i} x_i + b)

    where :math:`y` is output, :math:`x` is input, :math:`b` is bias,
    and :math:`f` is activation function.

    The example usage is:

    ..  code-block:: python

        addto = addto_layer(input=[layer1, layer2],
                            act=ReluActivation(),
                            bias_attr=False)

    This layer just simply add all input layers together, then activate the sum
    inputs. Each input of this layer should be the same size, which is also the
    output size of this layer.

C
caoying03 已提交
2886 2887 2888
    There is no weight matrix for each input, because it just a simple add
    operation. If you want a complicated operation before add, please use
    mixed_layer.
Z
zhangjinchao01 已提交
2889 2890

    It is a very good way to set dropout outside the layers. Since not all
C
caoying03 已提交
2891 2892
    PaddlePaddle layer support dropout, you can add an add_to layer, set
    dropout here.
Z
zhangjinchao01 已提交
2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906
    Please refer to dropout_layer for details.

    :param name: Layer name.
    :type name: basestring
    :param input: Input layers. It could be a LayerOutput or list/tuple of
                 LayerOutput.
    :type input: LayerOutput|list|tuple
    :param act: Activation Type, default is tanh.
    :type act: BaseActivation
    :param bias_attr: Bias attribute. If False, means no bias. None is default
                      bias.
    :type bias_attr: ParameterAttribute|bool
    :param layer_attr: Extra Layer attribute.
    :type layer_attr: ExtraLayerAttribute
D
dangqingqing 已提交
2907
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
2908 2909 2910 2911 2912 2913
    :rtype: LayerOutput
    """
    num_filters = None
    if isinstance(input, LayerOutput):
        input = [input]

2914
    assert isinstance(input, collections.Sequence)
Z
zhangjinchao01 已提交
2915 2916 2917 2918 2919 2920 2921
    ipts_for_layer = []
    for each_input in input:
        assert isinstance(each_input, LayerOutput)
        ipts_for_layer.append(Input(each_input.name))
        if each_input.num_filters is not None:
            num_filters = each_input.num_filters

X
xuwei06 已提交
2922
    l = Layer(
Q
qijun 已提交
2923 2924 2925
        name=name,
        type=LayerType.ADDTO_LAYER,
        inputs=ipts_for_layer,
Z
zhangjinchao01 已提交
2926 2927
        bias=ParamAttr.to_bias(bias_attr),
        active_type=act.name,
Q
qijun 已提交
2928
        **ExtraLayerAttribute.to_kwargs(layer_attr))
2929

Q
qijun 已提交
2930 2931 2932 2933 2934 2935 2936
    return LayerOutput(
        name,
        LayerType.ADDTO_LAYER,
        parents=input,
        activation=act,
        num_filters=num_filters,
        size=l.config.size)
Z
zhangjinchao01 已提交
2937 2938 2939 2940 2941


@wrap_act_default(act=IdentityActivation())
@wrap_name_default("concat")
@layer_support()
2942
def concat_layer(input, act=None, name=None, layer_attr=None, bias_attr=None):
Z
zhangjinchao01 已提交
2943 2944 2945 2946
    """
    Concat all input vector into one huge vector.
    Inputs can be list of LayerOutput or list of projection.

2947 2948 2949 2950 2951 2952
    The example usage is:

    ..  code-block:: python

        concat = concat_layer(input=[layer1, layer2])

Z
zhangjinchao01 已提交
2953 2954 2955
    :param name: Layer name.
    :type name: basestring
    :param input: input layers or projections
2956
    :type input: list|tuple|collections.Sequence
Z
zhangjinchao01 已提交
2957 2958 2959 2960
    :param act: Activation type.
    :type act: BaseActivation
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
D
dangqingqing 已提交
2961
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
2962 2963 2964 2965 2966 2967 2968 2969
    :rtype: LayerOutput
    """

    if isinstance(input, LayerOutput):
        input = [input]
    elif isinstance(input, Projection):
        input = [input]
    else:
2970
        assert isinstance(input, collections.Sequence)
Z
zhangjinchao01 已提交
2971 2972

    def __is_type__(o, tp):
2973
        if not isinstance(o, collections.Sequence):
Z
zhangjinchao01 已提交
2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994
            if o == tp:
                return True
            elif len(o.__bases__) == 0:
                return False
            else:
                for bs in o.__bases__:
                    if __is_type__(bs, tp):
                        return True
                return False
        else:
            tmp = map(lambda _x: __is_type__(_x, tp), o)
            a = tmp[0]
            for b in tmp[1:]:
                assert a == b
            return a

    def __reduce_concat_type__(a, b):
        assert __is_type__([a, b], Projection) or __is_type__([a, b],
                                                              LayerOutput)
        return a

Q
qijun 已提交
2995 2996
    is_concat_layer = __is_type__(
        reduce(__reduce_concat_type__, map(type, input)), LayerOutput)
Z
zhangjinchao01 已提交
2997

Q
qijun 已提交
2998 2999
    layer_type = (LayerType.CONCAT_LAYER
                  if is_concat_layer else LayerType.CONCAT_PROJ_LAYER)
Z
zhangjinchao01 已提交
3000

3001 3002
    if layer_type == LayerType.CONCAT_LAYER:
        assert not bias_attr
3003

3004
    layer = Layer(
Q
qijun 已提交
3005 3006
        name=name,
        type=layer_type,
Z
zhangjinchao01 已提交
3007 3008
        inputs=[x.name for x in input] if is_concat_layer else input,
        active_type=act.name,
3009
        bias=ParamAttr.to_bias(bias_attr),
Q
qijun 已提交
3010
        **ExtraLayerAttribute.to_kwargs(layer_attr))
Z
zhangjinchao01 已提交
3011

3012
    sz = layer.config.size
Z
zhangjinchao01 已提交
3013

Q
qijun 已提交
3014 3015 3016 3017 3018 3019 3020 3021
    return LayerOutput(
        name,
        layer_type=layer_type,
        parents=input if is_concat_layer else [x.origin for x in input],
        activation=act,
        size=sz)


3022 3023
@wrap_name_default("seqconcat")
@wrap_act_default(act=IdentityActivation())
3024
@wrap_bias_attr_default(has_bias=False)
3025 3026 3027 3028 3029
@layer_support()
def seq_concat_layer(a, b, act=None, name=None, layer_attr=None,
                     bias_attr=None):
    """
    Concat sequence a with sequence b.
3030

3031
    Inputs:
X
xuwei06 已提交
3032
      - a = [a1, a2, ..., am]
3033
      - b = [b1, b2, ..., bn]
3034

X
xuwei06 已提交
3035 3036 3037 3038
    Output: [a1, ..., am, b1, ..., bn]

    Note that the above computation is for one sample. Multiple samples are
    processed in one batch.
3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055

    The example usage is:

    ..  code-block:: python

        concat = seq_concat_layer(a=layer1, b=layer2)

    :param name: Layer name.
    :type name: basestring
    :param a: input sequence layer
    :type a: LayerOutput
    :param b: input sequence layer
    :type b: LayerOutput
    :param act: Activation type.
    :type act: BaseActivation
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
3056 3057 3058 3059
    :param bias_attr: The Bias Attribute. If no bias, then pass False or
                      something not type of ParameterAttribute. None will get a
                      default Bias.
    :type bias_attr: ParameterAttribute or None or bool
3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080
    :return: LayerOutput object.
    :rtype: LayerOutput
    """
    assert isinstance(a, LayerOutput) and isinstance(b, LayerOutput)
    assert a.size == b.size
    Layer(
        name=name,
        type=LayerType.SEQUENCE_CONCAT_LAYER,
        inputs=[a.name, b.name],
        active_type=act.name,
        bias=ParamAttr.to_bias(bias_attr),
        **ExtraLayerAttribute.to_kwargs(layer_attr))

    return LayerOutput(
        name,
        layer_type=LayerType.SEQUENCE_CONCAT_LAYER,
        parents=[a, b],
        activation=act,
        size=a.size)


3081
@wrap_name_default("memory", "memory_name")
Q
qijun 已提交
3082 3083
def memory(name,
           size,
3084
           memory_name=None,
Q
qijun 已提交
3085 3086 3087 3088
           is_seq=False,
           boot_layer=None,
           boot_bias=None,
           boot_bias_active_type=None,
Z
zhangjinchao01 已提交
3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108
           boot_with_const_id=None):
    """
    The memory layers is a layer cross each time step. Reference this output
    as previous time step layer :code:`name` 's output.

    The default memory is zero in first time step, previous time step's
    output in the rest time steps.

    If boot_bias, the first time step value is this bias and
    with activation.

    If boot_with_const_id, then the first time stop is a IndexSlot, the
    Arguments.ids()[0] is this :code:`cost_id`.

    If boot_layer is not null, the memory is just the boot_layer's output.
    Set :code:`is_seq` is true boot layer is sequence.

    The same name layer in recurrent group will set memory on each time
    step.

3109 3110 3111 3112 3113 3114 3115 3116 3117
    .. code-block:: python

       mem = memory(size=256, name='state')
       state = fc_layer(input=mem, size=256, name='state')

    If you do not want to specify the name, you can equivalently use set_input()
    to specify the layer needs to be remembered as the following:

    .. code-block:: python
L
Liu Yiqun 已提交
3118

3119 3120 3121 3122 3123 3124 3125
       mem = memory(size=256)
       state = fc_layer(input=mem, size=256)
       mem.set_input(mem)

    :param name: the name of the layer which this memory remembers.
                 If name is None, user should call set_input() to specify the
                 name of the layer which this memory remembers.
Z
zhangjinchao01 已提交
3126 3127 3128
    :type name: basestring
    :param size: size of memory.
    :type size: int
3129 3130 3131
    :param memory_name: the name of the memory.
                        It is ignored when name is provided.
    :type memory_name: basestring
3132
    :param is_seq: DEPRECATED. is sequence for boot_layer
Z
zhangjinchao01 已提交
3133 3134 3135 3136 3137 3138 3139 3140 3141
    :type is_seq: bool
    :param boot_layer: boot layer of memory.
    :type boot_layer: LayerOutput|None
    :param boot_bias: boot layer's bias
    :type boot_bias: ParameterAttribute|None
    :param boot_bias_active_type: boot layer's active type.
    :type boot_bias_active_type: BaseActivation
    :param boot_with_const_id: boot layer's id.
    :type boot_with_const_id: int
D
dangqingqing 已提交
3142
    :return: LayerOutput object which is a memory.
Z
zhangjinchao01 已提交
3143 3144 3145 3146 3147 3148 3149 3150 3151 3152
    :rtype: LayerOutput
    """
    if boot_bias_active_type is None:
        boot_bias_active_type = LinearActivation()

    assert boot_bias is None or isinstance(boot_bias, ParameterAttribute)
    if isinstance(boot_bias, ParameterAttribute):
        boot_bias = ParamAttr.to_bias(boot_bias)

    assert boot_layer is None or isinstance(boot_layer, LayerOutput)
3153 3154
    if name is not None:
        memory_name = None
Z
zhangjinchao01 已提交
3155

3156 3157 3158 3159 3160 3161 3162 3163
    memory_name = Memory(
        name,
        size,
        boot_layer=boot_layer.name if boot_layer is not None else None,
        boot_bias=boot_bias,
        boot_bias_active_type=boot_bias_active_type.name,
        boot_with_const_id=boot_with_const_id,
        memory_name=memory_name)
Q
qijun 已提交
3164 3165

    lout = LayerOutput(
3166
        name=memory_name,
Q
qijun 已提交
3167 3168 3169
        size=size,
        layer_type=LayerType.MEMORY,
        parents=[boot_layer] if boot_layer is not None else None)
Z
zhangjinchao01 已提交
3170 3171 3172 3173
    return lout


@wrap_bias_attr_default()
Q
qijun 已提交
3174 3175
@wrap_act_default(
    param_names=['gate_act', 'state_act'], act=SigmoidActivation())
Z
zhangjinchao01 已提交
3176 3177 3178
@wrap_act_default(act=TanhActivation())
@wrap_name_default('lstm_step')
@layer_support()
Q
qijun 已提交
3179 3180
def lstm_step_layer(input,
                    state,
3181
                    size=None,
Q
qijun 已提交
3182 3183 3184 3185 3186 3187
                    act=None,
                    name=None,
                    gate_act=None,
                    state_act=None,
                    bias_attr=None,
                    layer_attr=None):
Z
zhangjinchao01 已提交
3188
    """
3189 3190
    LSTM Step Layer. This function is used only in recurrent_group.
    The lstm equations are shown as follows.
Z
zhangjinchao01 已提交
3191 3192 3193

    ..  math::

3194
        i_t & = \\sigma(W_{x_i}x_{t} + W_{h_i}h_{t-1} + W_{c_i}c_{t-1} + b_i)
Z
zhangjinchao01 已提交
3195

3196
        f_t & = \\sigma(W_{x_f}x_{t} + W_{h_f}h_{t-1} + W_{c_f}c_{t-1} + b_f)
Z
zhangjinchao01 已提交
3197

3198
        c_t & = f_tc_{t-1} + i_t tanh (W_{x_c}x_t+W_{h_c}h_{t-1} + b_c)
Z
zhangjinchao01 已提交
3199

3200
        o_t & = \\sigma(W_{x_o}x_{t} + W_{h_o}h_{t-1} + W_{c_o}c_t + b_o)
Z
zhangjinchao01 已提交
3201

L
luotao02 已提交
3202
        h_t & = o_t tanh(c_t)
Z
zhangjinchao01 已提交
3203 3204


L
luotao02 已提交
3205
    The input of lstm step is :math:`Wx_t + Wh_{t-1}`, and user should use
Z
zhangjinchao01 已提交
3206
    :code:`mixed_layer` and :code:`full_matrix_projection` to calculate these
3207
    input vectors.
Z
zhangjinchao01 已提交
3208 3209 3210 3211 3212 3213 3214 3215 3216 3217

    The state of lstm step is :math:`c_{t-1}`. And lstm step layer will do

    ..  math::

        i_t = \\sigma(input + W_{ci}c_{t-1} + b_i)

        ...


3218 3219
    This layer has two outputs. Default output is :math:`h_t`. The other
    output is :math:`o_t`, whose name is 'state' and can use
Z
zhangjinchao01 已提交
3220 3221 3222 3223
    :code:`get_output_layer` to extract this output.

    :param name: Layer's name.
    :type name: basestring
3224 3225
    :param size: Layer's size. NOTE: lstm layer's size, should be equal to
                 :code:`input.size/4`, and should be equal to
Z
zhangjinchao01 已提交
3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243
                 :code:`state.size`.
    :type size: int
    :param input: input layer. :math:`Wx_t + Wh_{t-1}`
    :type input: LayerOutput
    :param state: State Layer. :math:`c_{t-1}`
    :type state: LayerOutput
    :param act: Activation type. Default is tanh
    :type act: BaseActivation
    :param gate_act: Gate Activation Type. Default is sigmoid, and should
                          be sigmoid only.
    :type gate_act: BaseActivation
    :param state_act: State Activation Type. Default is sigmoid, and should
                           be sigmoid only.
    :type state_act: BaseActivation
    :param bias_attr: Bias Attribute.
    :type bias_attr: ParameterAttribute
    :param layer_attr: layer's extra attribute.
    :type layer_attr: ExtraLayerAttribute
D
dangqingqing 已提交
3244
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
3245 3246
    :rtype: LayerOutput
    """
3247 3248 3249

    assert size is None or state.size == size
    size = state.size
Z
zhangjinchao01 已提交
3250 3251 3252 3253 3254 3255 3256
    Layer(
        name=name,
        type=LayerType.LSTM_STEP_LAYER,
        active_type=act.name,
        active_gate_type=gate_act.name,
        active_state_type=state_act.name,
        bias=ParamAttr.to_bias(bias_attr),
3257
        size=state.size,
Q
qijun 已提交
3258 3259
        inputs=[input.name, state.name],
        **ExtraLayerAttribute.to_kwargs(layer_attr))
Z
zhangjinchao01 已提交
3260

Q
qijun 已提交
3261 3262 3263 3264 3265 3266 3267
    return LayerOutput(
        name=name,
        layer_type=LayerType.LSTM_STEP_LAYER,
        parents=[input, state],
        activation=act,
        size=size,
        outputs=['default', 'state'])
Z
zhangjinchao01 已提交
3268 3269 3270


@wrap_bias_attr_default()
W
wangyang59 已提交
3271
@wrap_param_attr_default()
Q
qijun 已提交
3272
@wrap_act_default(param_names=['gate_act'], act=SigmoidActivation())
Z
zhangjinchao01 已提交
3273 3274 3275
@wrap_act_default(act=TanhActivation())
@wrap_name_default('gru_step')
@layer_support()
Q
qijun 已提交
3276 3277 3278 3279 3280 3281 3282
def gru_step_layer(input,
                   output_mem,
                   size=None,
                   act=None,
                   name=None,
                   gate_act=None,
                   bias_attr=None,
W
wangyang59 已提交
3283
                   param_attr=None,
Q
qijun 已提交
3284
                   layer_attr=None):
Z
zhangjinchao01 已提交
3285 3286 3287 3288 3289 3290 3291 3292 3293 3294
    """

    :param input:
    :type input: LayerOutput
    :param output_mem:
    :param size:
    :param act:
    :param name:
    :param gate_act:
    :param bias_attr:
3295 3296
    :param param_attr: the parameter_attribute for transforming the output_mem
                       from previous step.
Z
zhangjinchao01 已提交
3297
    :param layer_attr:
D
dangqingqing 已提交
3298
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
3299 3300 3301 3302 3303 3304 3305 3306
    :rtype: LayerOutput
    """
    assert input.size % 3 == 0
    if size is None:
        size = input.size / 3
    Layer(
        name=name,
        type=LayerType.GRU_STEP_LAYER,
3307 3308 3309 3310
        # The parameter here is for transforming the output_mem. The input has
        # already been transformed outside this module so it does not need
        # parameter associated with it.
        # The parameter here is instead grouped with input is due to
3311
        # backward model compatibility.
3312
        inputs=[Input(input.name, **param_attr.attr), output_mem.name],
Z
zhangjinchao01 已提交
3313 3314 3315 3316
        bias=ParamAttr.to_bias(bias_attr),
        size=size,
        active_type=act.name,
        active_gate_type=gate_act.name,
Q
qijun 已提交
3317
        **ExtraAttr.to_kwargs(layer_attr))
Z
zhangjinchao01 已提交
3318
    return LayerOutput(
Q
qijun 已提交
3319 3320
        name=name,
        layer_type=LayerType.GRU_STEP_LAYER,
Z
zhangjinchao01 已提交
3321
        parents=[input, output_mem],
Q
qijun 已提交
3322 3323
        size=size,
        activation=act)
Z
zhangjinchao01 已提交
3324 3325


Y
Yu Yang 已提交
3326 3327 3328 3329
@wrap_bias_attr_default()
@wrap_param_attr_default()
@wrap_act_default(param_names=['gate_act'], act=SigmoidActivation())
@wrap_act_default(act=TanhActivation())
Q
qijun 已提交
3330
@wrap_name_default('gru_step_naive')
Y
Yu Yang 已提交
3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397
@layer_support(ERROR_CLIPPING, DROPOUT)
def gru_step_naive_layer(input,
                         output_mem,
                         size=None,
                         name=None,
                         act=None,
                         gate_act=None,
                         bias_attr=None,
                         param_attr=None,
                         layer_attr=None):
    """
    GRU Step Layer, but using MixedLayer to generate. It support ERROR_CLIPPING
    and DROPOUT.

    :param input:
    :param output_mem:
    :param size:
    :param name:
    :param act:
    :param gate_act:
    :param bias_attr:
    :param param_attr:
    :param layer_attr:
    :return:
    """
    if input.size % 3 != 0:
        raise ValueError("GruStep input size must be divided by 3")
    if size is None:
        size = input.size / 3

    def __gate__(gate_name, offset):
        with mixed_layer(
                name=name + "_" + gate_name,
                size=size,
                layer_attr=layer_attr,
                bias_attr=bias_attr,
                act=gate_act) as gate:
            gate += identity_projection(input=input, offset=offset)
            gate += full_matrix_projection(
                input=output_mem, param_attr=param_attr)
        return gate

    update_gate = __gate__("update", 0)
    reset_gate = __gate__("reset", size)

    with mixed_layer(
            name=name + "_reset_output", bias_attr=False) as reset_output:
        reset_output += dotmul_operator(a=output_mem, b=reset_gate)

    with mixed_layer(
            name=name + "_output_candidate",
            size=size,
            layer_attr=layer_attr,
            bias_attr=bias_attr,
            act=act) as output_candidate:
        output_candidate += identity_projection(input=input, offset=2 * size)
        output_candidate += full_matrix_projection(
            input=reset_output, param_attr=param_attr)

    with mixed_layer(name=name) as output:
        output += identity_projection(output_mem)
        output += dotmul_operator(a=output_mem, b=update_gate, scale=-1.0)
        output += dotmul_operator(a=output_candidate, b=update_gate)

    return output


Z
zhangjinchao01 已提交
3398 3399 3400 3401
@wrap_name_default()
@layer_support()
def get_output_layer(input, arg_name, name=None, layer_attr=None):
    """
C
caoying03 已提交
3402 3403 3404 3405
    Get layer's output by name. In PaddlePaddle, a layer might return multiple
    values, but returns one layer's output. If the user wants to use another
    output besides the default one, please use get_output_layer first to get
    the output from input.
Z
zhangjinchao01 已提交
3406 3407 3408 3409 3410 3411 3412 3413 3414

    :param name: Layer's name.
    :type name: basestring
    :param input: get output layer's input. And this layer should contains
                   multiple outputs.
    :type input: LayerOutput
    :param arg_name: Output name from input.
    :type arg_name: basestring
    :param layer_attr: Layer's extra attribute.
D
dangqingqing 已提交
3415
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
3416 3417 3418 3419 3420 3421 3422
    :rtype: LayerOutput
    """
    # GetOutputLayer
    assert arg_name in input.outputs, 'Get Output From an not existed input.' \
                                      ' The get output name is %s, which not' \
                                      ' in %s' % (
                                          arg_name, ",".join(input.outputs))
Q
qijun 已提交
3423 3424 3425 3426 3427 3428 3429
    Layer(
        name=name,
        type=LayerType.GET_OUTPUT_LAYER,
        inputs=[Input(
            input.name, input_layer_argument=arg_name)],
        size=input.size,
        **ExtraLayerAttribute.to_kwargs(layer_attr))
Z
zhangjinchao01 已提交
3430

Q
qijun 已提交
3431 3432 3433 3434 3435
    return LayerOutput(
        name=name,
        layer_type=LayerType.GET_OUTPUT_LAYER,
        parents=[input],
        size=input.size)
Z
zhangjinchao01 已提交
3436 3437 3438 3439 3440 3441 3442


@wrap_name_default()
@wrap_act_default()
@wrap_bias_attr_default()
@wrap_param_attr_default()
@layer_support()
Q
qijun 已提交
3443 3444 3445 3446 3447 3448 3449
def recurrent_layer(input,
                    act=None,
                    bias_attr=None,
                    param_attr=None,
                    name=None,
                    reverse=False,
                    layer_attr=None):
Z
zhangjinchao01 已提交
3450
    """
3451 3452
    Simple recurrent unit layer. It is just a fully connect layer through both
    time and neural network.
Z
zhangjinchao01 已提交
3453

3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480
    For each sequence [start, end] it performs the following computation\:

    ..  math::

        out_{i} = act(in_{i})     \\      \\      \\text{for} \\ i = start \\\\
        out_{i} = act(in_{i} + out_{i-1} * W) \\ \\ \\text{for} \\ start < i <= end

    If reversed is true, the order is reversed\:

    ..  math::

        out_{i} = act(in_{i})           \\    \\   \\text{for} \\ i = end  \\\\
        out_{i} = act(in_{i} + out_{i+1} * W) \\ \\ \\text{for} \\ start <= i < end


    :param input: Input Layer
    :type input: LayerOutput
    :param act: activation.
    :type act: BaseActivation
    :param bias_attr: bias attribute.
    :type bias_attr: ParameterAttribute
    :param param_attr: parameter attribute.
    :type param_attr: ParameterAttribute
    :param name: name of the layer
    :type name: basestring
    :param layer_attr: Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
D
dangqingqing 已提交
3481
    :return: LayerOutput object.
3482
    :rtype: LayerOutput
Z
zhangjinchao01 已提交
3483
    """
Q
qijun 已提交
3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498
    Layer(
        name=name,
        type=LayerType.RECURRENT_LAYER,
        inputs=Input(input.name, **param_attr.attr),
        active_type=act.name,
        bias=ParamAttr.to_bias(bias_attr),
        reversed=reverse,
        **ExtraAttr.to_kwargs(layer_attr))
    return LayerOutput(
        name=name,
        layer_type=LayerType.RECURRENT_LAYER,
        parents=[input],
        size=input.size,
        activation=act,
        reverse=reverse)
Z
zhangjinchao01 已提交
3499 3500 3501 3502 3503 3504


class StaticInput(object):
    """
    StaticInput is only used in recurrent_group which defines a read-only memory
    that can be a sequence or non-sequence.
3505 3506
    :param size: DEPRECATED
    :param is_seq: DEPRECATED
Z
zhangjinchao01 已提交
3507
    """
3508

Z
zhangjinchao01 已提交
3509 3510 3511
    def __init__(self, input, is_seq=False, size=None):
        assert isinstance(input, LayerOutput)
        self.input = input
3512
        assert input.size is not None
Z
zhangjinchao01 已提交
3513
        if size is not None:
3514
            assert input.size == size
Z
zhangjinchao01 已提交
3515 3516


3517
def SubsequenceInput(input):
Z
zhangjinchao01 已提交
3518
    """
3519
    DEPRECATED.
Z
zhangjinchao01 已提交
3520 3521 3522 3523 3524 3525 3526 3527
    Input sequence has sub-sequence, used in recurrent_group.

    The example usage is:

    .. code-block:: python

       input = SubsequenceInput(layer)
    """
3528
    return input
Z
zhangjinchao01 已提交
3529 3530 3531


@wrap_name_default("recurrent_group")
L
Luo Tao 已提交
3532 3533 3534 3535 3536
def recurrent_group(step,
                    input,
                    reverse=False,
                    name=None,
                    targetInlink=None,
L
Luo Tao 已提交
3537
                    is_generating=False):
Z
zhangjinchao01 已提交
3538
    """
C
caoying03 已提交
3539 3540 3541 3542 3543
    Recurrent layer group is an extremely flexible recurrent unit in
    PaddlePaddle. As long as the user defines the calculation done within a
    time step, PaddlePaddle will iterate such a recurrent calculation over
    sequence input. This is extremely usefull for attention based model, or
    Neural Turning Machine like models.
Z
zhangjinchao01 已提交
3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587

    The basic usage (time steps) is:

    .. code-block:: python

       def step(input):
           output = fc_layer(input=layer,
                             size=1024,
                             act=LinearActivation(),
                             bias_attr=False)
           return output

       group = recurrent_group(input=layer,
                               step=step)

    You can see following configs for further usages:

    - time steps: lstmemory_group, paddle/gserver/tests/sequence_layer_group.conf, \
                  demo/seqToseq/seqToseq_net.py
    - sequence steps: paddle/gserver/tests/sequence_nest_layer_group.conf

    :param step: recurrent one time step function.The input of this function is
                 input of the group. The return of this function will be
                 recurrent group's return value.

                 The recurrent group scatter a sequence into time steps. And
                 for each time step, will invoke step function, and return
                 a time step result. Then gather each time step of output into
                 layer group's output.

    :type step: callable

    :param name: recurrent_group's name.
    :type name: basestring

    :param input: Input links array.

                  LayerOutput will be scattered into time steps.
                  SubsequenceInput will be scattered into sequence steps.
                  StaticInput will be imported to each time step, and doesn't change
                  through time. It's a mechanism to access layer outside step function.

    :type input: LayerOutput|StaticInput|SubsequenceInput|list|tuple

3588 3589
    :param reverse: If reverse is set true, the recurrent unit will process the
                    input sequence in a reverse order.
Z
zhangjinchao01 已提交
3590
    :type reverse: bool
3591

3592 3593
    :param targetInlink: DEPRECATED.
                         The input layer which share info with layer group's output
3594 3595 3596 3597 3598 3599 3600 3601 3602

                         Param input specifies multiple input layers. For
                         SubsequenceInput inputs, config should assign one input
                         layer that share info(the number of sentences and the number
                         of words in each sentence) with all layer group's outputs.
                         targetInlink should be one of the layer group's input.

    :type targetInlink: LayerOutput|SubsequenceInput

L
Luo Tao 已提交
3603
    :param is_generating: If is generating, none of input type should be LayerOutput;
3604
                          else, for training or testing, one of the input type must
L
Luo Tao 已提交
3605
                          be LayerOutput.
L
Luo Tao 已提交
3606

L
Liu Yiqun 已提交
3607
    :type is_generating: bool
3608

D
dangqingqing 已提交
3609
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
3610 3611 3612 3613 3614
    :rtype: LayerOutput
    """
    model_type('recurrent_nn')

    def is_single_input(x):
3615
        return isinstance(x, LayerOutput) or isinstance(x, StaticInput)
Z
zhangjinchao01 已提交
3616 3617 3618

    if is_single_input(input):
        input = [input]
3619
    assert isinstance(input, collections.Sequence)
Z
zhangjinchao01 已提交
3620 3621

    def is_in_links(x):
3622
        return isinstance(x, LayerOutput)
Z
zhangjinchao01 已提交
3623 3624 3625 3626

    in_links = filter(is_in_links, input)

    RecurrentLayerGroupWithoutOutLinksBegin(
Q
qijun 已提交
3627
        name=name,
3628 3629
        in_links=map(lambda x: x.name, in_links),
        seq_reversed=reverse)
Z
zhangjinchao01 已提交
3630
    in_args = []
3631
    has_LayerOutput = False
Z
zhangjinchao01 已提交
3632 3633 3634 3635
    for each_input in input:
        assert is_single_input(each_input)
        if isinstance(each_input, LayerOutput):
            in_args.append(each_input)
3636
            has_LayerOutput = True
3637
        else:  # StaticInput
Z
zhangjinchao01 已提交
3638
            mem_name = "__%s_memory__" % each_input.input.name
Q
qijun 已提交
3639
            mem = memory(
3640
                name=None,
Q
qijun 已提交
3641 3642
                size=each_input.input.size,
                boot_layer=each_input.input)
3643
            mem.set_input(mem)
Z
zhangjinchao01 已提交
3644 3645
            in_args.append(mem)

L
Luo Tao 已提交
3646
    assert (is_generating != has_LayerOutput)
L
Luo Tao 已提交
3647

Z
zhangjinchao01 已提交
3648 3649 3650 3651 3652 3653 3654
    layer_outs = step(*in_args)

    if isinstance(layer_outs, LayerOutput):
        layer_outs = [layer_outs]

    for ot in layer_outs:
        assert isinstance(ot, LayerOutput)
3655
        ot.reverse = reverse
3656
        RecurrentLayerGroupSetOutLink(ot.name)
Z
zhangjinchao01 已提交
3657 3658 3659

    RecurrentLayerGroupEnd(name=name)

X
xuwei06 已提交
3660 3661 3662 3663 3664
    for layer_out in layer_outs:
        # Thee previous full_name is the name is the rnn group
        # We need a full_name outside the rnn group
        layer_out.full_name = MakeLayerNameInSubmodel(layer_out.name)

Z
zhangjinchao01 已提交
3665 3666 3667 3668 3669
    if len(layer_outs) == 1:
        return layer_outs[0]
    else:
        return layer_outs

3670

Z
zhangjinchao01 已提交
3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687
class BaseGeneratedInput(object):
    def __init__(self):
        self.bos_id = None
        self.eos_id = None

    def before_real_step(self):
        raise NotImplementedError()

    def after_real_step(self, *args):
        raise NotImplementedError()


class GeneratedInput(BaseGeneratedInput):
    def after_real_step(self, input):
        return maxid_layer(input=input, name='__beam_search_predict__')

    def before_real_step(self):
Q
qijun 已提交
3688 3689 3690 3691 3692 3693 3694 3695 3696
        predict_id = memory(
            name='__beam_search_predict__',
            size=self.size,
            boot_with_const_id=self.bos_id)

        trg_emb = embedding_layer(
            input=predict_id,
            size=self.embedding_size,
            param_attr=ParamAttr(name=self.embedding_name))
Z
zhangjinchao01 已提交
3697 3698 3699
        return trg_emb

    def __init__(self, size, embedding_name, embedding_size):
3700
        super(GeneratedInput, self).__init__()
Z
zhangjinchao01 已提交
3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723
        self.size = size
        self.embedding_name = embedding_name
        self.embedding_size = embedding_size


@wrap_name_default()
def maxid_layer(input, name=None, layer_attr=None):
    """
    A layer for finding the id which has the maximal value for each sample.
    The result is stored in output.ids.

    The example usage is:

    .. code-block:: python

       maxid = maxid_layer(input=layer)

    :param input: Input layer name.
    :type input: LayerOutput
    :param name: Layer name.
    :type name: basestring
    :param layer_attr: extra layer attributes.
    :type layer_attr: ExtraLayerAttribute.
D
dangqingqing 已提交
3724
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
3725 3726 3727 3728
    :rtype: LayerOutput
    """

    assert isinstance(input, LayerOutput)
Q
qijun 已提交
3729 3730 3731 3732 3733 3734 3735 3736 3737 3738
    l = Layer(
        name=name,
        type='maxid',
        inputs=[input.name],
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name=name,
        layer_type=LayerType.MAXID_LAYER,
        parents=[input],
        size=l.config.size)
Z
zhangjinchao01 已提交
3739

3740

H
Haonan 已提交
3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766
@wrap_name_default()
def out_prod_layer(input1, input2, name=None, layer_attr=None):
    """
    A layer for computing the outer product of two vectors
    The result is a matrix of size(input1) x size(input2)

    The example usage is:

    .. code-block:: python

       out_prod = out_prod_layer(input1=vec1, input2=vec2)

    :param name: Layer name.
    :type name: basestring
    :param input1: The first input layer name.
    :type input: LayerOutput
    :param input2: The second input layer name.
    :type input2: LayerOutput
    :param layer_attr: extra layer attributes.
    :type layer_attr: ExtraLayerAttribute.
    :return: LayerOutput object.
    :rtype: LayerOutput
    """

    assert isinstance(input1, LayerOutput)
    assert isinstance(input2, LayerOutput)
Q
qijun 已提交
3767 3768 3769 3770 3771 3772 3773 3774 3775 3776
    l = Layer(
        name=name,
        type=LayerType.OUT_PROD_LAYER,
        inputs=[input1.name, input2.name],
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name=name,
        layer_type=LayerType.OUT_PROD_LAYER,
        parents=[input1, input2],
        size=l.config.size)
3777

Z
zhangjinchao01 已提交
3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793

@wrap_name_default()
def eos_layer(input, eos_id, name=None, layer_attr=None):
    """
    A layer for checking EOS for each sample:
    - output_id = (input_id == conf.eos_id)

    The result is stored in output\_.ids.
    It is used by recurrent layer group.

    The example usage is:

    .. code-block:: python

       eos = eos_layer(input=layer, eos_id=id)

L
luotao02 已提交
3794 3795
    :param name: Layer name.
    :type name: basestring
Z
zhangjinchao01 已提交
3796 3797 3798 3799 3800 3801
    :param input: Input layer name.
    :type input: LayerOutput
    :param eos_id: end id of sequence
    :type eos_id: int
    :param layer_attr: extra layer attributes.
    :type layer_attr: ExtraLayerAttribute.
D
dangqingqing 已提交
3802
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
3803 3804
    :rtype: LayerOutput
    """
Q
qijun 已提交
3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815
    l = Layer(
        name=name,
        type=LayerType.EOSID_LAYER,
        eos_id=eos_id,
        inputs=[input.name],
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name=name,
        layer_type=LayerType.EOSID_LAYER,
        parents=[input],
        size=l.config.size)
Z
zhangjinchao01 已提交
3816 3817 3818


@wrap_name_default()
Q
qijun 已提交
3819 3820 3821 3822 3823 3824 3825
def beam_search(step,
                input,
                bos_id,
                eos_id,
                beam_size,
                max_length=500,
                name=None,
Z
zhangjinchao01 已提交
3826
                num_results_per_sample=None):
3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837
    """
    Beam search is a heuristic search algorithm used in sequence generation.
    It explores a graph by expanding the most promising nodes in a limited set
    to maintain tractability.

    The example usage is:

    .. code-block:: python

        def rnn_step(input):
            last_time_step_output = memory(name='rnn', size=512)
3838
            with mixed_layer(size=512, name='rnn') as simple_rnn:
3839 3840 3841 3842
                simple_rnn += full_matrix_projection(input)
                simple_rnn += last_time_step_output
            return simple_rnn

3843 3844 3845 3846 3847
        generated_word_embedding = GeneratedInput(
                               size=target_dictionary_dim,
                               embedding_name="target_language_embedding",
                               embedding_size=word_vector_dim)

3848 3849
        beam_gen = beam_search(name="decoder",
                               step=rnn_step,
3850 3851
                               input=[StaticInput(encoder_last),
                                      generated_word_embedding],
3852 3853
                               bos_id=0,
                               eos_id=1,
3854
                               beam_size=5)
3855 3856 3857 3858 3859 3860 3861 3862 3863

    Please see the following demo for more details:

    - machine translation : demo/seqToseq/translation/gen.conf \
                            demo/seqToseq/seqToseq_net.py

    :param name: Name of the recurrent unit that generates sequences.
    :type name: base string
    :param step: A callable function that defines the calculation in a time
3864
                 step, and it is applied to sequences with arbitrary length by
3865 3866 3867 3868 3869
                 sharing a same set of weights.

                 You can refer to the first parameter of recurrent_group, or
                 demo/seqToseq/seqToseq_net.py for more details.
    :type step: callable
3870 3871
    :param input: Input data for the recurrent unit, which should include the
                  previously generated words as a GeneratedInput object.
3872
    :type input: list
3873 3874 3875
    :param bos_id: Index of the start symbol in the dictionary. The start symbol
                   is a special token for NLP task, which indicates the
                   beginning of a sequence. In the generation task, the start
3876
                   symbol is essential, since it is used to initialize the RNN
3877 3878 3879 3880 3881 3882 3883 3884
                   internal state.
    :type bos_id: int
    :param eos_id: Index of the end symbol in the dictionary. The end symbol is
                   a special token for NLP task, which indicates the end of a
                   sequence. The generation process will stop once the end
                   symbol is generated, or a pre-defined max iteration number
                   is exceeded.
    :type eos_id: int
3885 3886
    :param max_length: Max generated sequence length.
    :type max_length: int
3887 3888 3889 3890 3891 3892 3893 3894 3895 3896
    :param beam_size: Beam search for sequence generation is an iterative search
                      algorithm. To maintain tractability, every iteration only
                      only stores a predetermined number, called the beam_size,
                      of the most promising next words. The greater the beam
                      size, the fewer candidate words are pruned.
    :type beam_size: int
    :param num_results_per_sample: Number of the generated results per input
                                  sequence. This number must always be less than
                                  beam size.
    :type num_results_per_sample: int
3897 3898
    :return: The generated word index.
    :rtype: LayerOutput
3899 3900
    """

Z
zhangjinchao01 已提交
3901 3902 3903 3904 3905
    if num_results_per_sample is None:
        num_results_per_sample = beam_size
    if num_results_per_sample > beam_size:
        logger.warning("num_results_per_sample should be less than beam_size")

Q
qijun 已提交
3906
    if isinstance(input, StaticInput) or isinstance(input, BaseGeneratedInput):
Z
zhangjinchao01 已提交
3907 3908 3909 3910 3911 3912
        input = [input]

    generated_input_index = -1

    real_input = []
    for i, each_input in enumerate(input):
3913 3914
        assert isinstance(each_input, StaticInput) or isinstance(
            each_input, BaseGeneratedInput)
Z
zhangjinchao01 已提交
3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929
        if isinstance(each_input, BaseGeneratedInput):
            assert generated_input_index == -1
            generated_input_index = i
        else:
            real_input.append(each_input)

    assert generated_input_index != -1

    gipt = input[generated_input_index]

    gipt.bos_id = bos_id
    gipt.eos_id = eos_id

    def __real_step__(*args):
        eos_name = "__%s_eos_layer__" % name
Q
qijun 已提交
3930 3931 3932 3933 3934 3935
        RecurrentLayerGroupSetGenerator(
            Generator(
                eos_layer_name=eos_name,
                max_num_frames=max_length,
                beam_size=beam_size,
                num_results_per_sample=num_results_per_sample))
Z
zhangjinchao01 已提交
3936 3937 3938 3939 3940 3941 3942 3943 3944

        args = list(args)
        args.insert(generated_input_index, gipt.before_real_step())

        predict = gipt.after_real_step(step(*args))

        eos_layer(input=predict, eos_id=eos_id, name=eos_name)
        return predict

Q
qijun 已提交
3945
    tmp = recurrent_group(
L
Luo Tao 已提交
3946 3947 3948 3949
        step=__real_step__,
        input=real_input,
        reverse=False,
        name=name,
L
Luo Tao 已提交
3950
        is_generating=True)
3951

Z
zhangjinchao01 已提交
3952 3953
    return tmp

Q
qijun 已提交
3954

3955 3956
def __cost_input__(input, label, weight=None):
    """
3957
    inputs and parents for cost layers.
3958 3959 3960 3961
    """
    ipts = [Input(input.name), Input(label.name)]
    parents = [input, label]
    if weight is not None:
3962
        assert weight.size == 1
3963 3964 3965
        ipts.append(Input(weight.name))
        parents.append(weight)
    return ipts, parents
3966

Z
zhangjinchao01 已提交
3967 3968

@wrap_name_default()
L
luotao1 已提交
3969
@layer_support()
3970
def mse_cost(input, label, weight=None, name=None, coeff=1.0, layer_attr=None):
Z
zhangjinchao01 已提交
3971
    """
L
Luo Tao 已提交
3972 3973 3974 3975
    mean squared error cost:

    ..  math::

L
Liu Yiqun 已提交
3976
        \\frac{1}{N}\sum_{i=1}^N(t_i-y_i)^2
Z
zhangjinchao01 已提交
3977 3978

    :param name: layer name.
3979
    :type name: basestring
Z
zhangjinchao01 已提交
3980
    :param input: Network prediction.
3981
    :type input: LayerOutput
Z
zhangjinchao01 已提交
3982
    :param label: Data label.
3983 3984 3985 3986
    :type label: LayerOutput
    :param weight: The weight affects the cost, namely the scale of cost.
                   It is an optional argument.
    :type weight: LayerOutput
3987 3988
    :param coeff: The coefficient affects the gradient in the backward.
    :type coeff: float
L
luotao1 已提交
3989 3990
    :param layer_attr: layer's extra attribute.
    :type layer_attr: ExtraLayerAttribute
D
dangqingqing 已提交
3991
    :return: LayerOutput object.
3992
    :rtype: LayerOutput
Z
zhangjinchao01 已提交
3993
    """
3994 3995
    ipts, parents = __cost_input__(input, label, weight)

Q
qijun 已提交
3996 3997 3998 3999
    Layer(
        inputs=ipts,
        type="square_error",
        name=name,
4000
        coeff=coeff,
Q
qijun 已提交
4001
        **ExtraLayerAttribute.to_kwargs(layer_attr))
X
xuwei06 已提交
4002
    return LayerOutput(name, LayerType.COST, parents=parents, size=1)
Z
zhangjinchao01 已提交
4003 4004


L
Luo Tao 已提交
4005 4006 4007
regression_cost = mse_cost


Z
zhangjinchao01 已提交
4008
@wrap_name_default("cost")
4009
@layer_support()
Q
qijun 已提交
4010 4011 4012 4013
def classification_cost(input,
                        label,
                        weight=None,
                        name=None,
4014
                        evaluator=classification_error_evaluator,
4015 4016
                        layer_attr=None,
                        coeff=1.):
Z
zhangjinchao01 已提交
4017 4018 4019 4020 4021 4022 4023 4024 4025
    """
    classification cost Layer.

    :param name: layer name.
    :type name: basestring
    :param input: input layer name. network output.
    :type input: LayerOutput
    :param label: label layer name. data_layer often.
    :type label: LayerOutput
4026 4027 4028
    :param weight: The weight affects the cost, namely the scale of cost.
                   It is an optional argument.
    :type weight: LayerOutput
Z
zhangjinchao01 已提交
4029
    :param evaluator: Evaluator method.
4030 4031
    :param layer_attr: layer's extra attribute.
    :type layer_attr: ExtraLayerAttribute
4032 4033
    :param coeff: The coefficient affects the gradient in the backward.
    :type coeff: float
D
dangqingqing 已提交
4034
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
4035 4036 4037 4038 4039
    :rtype: LayerOutput
    """
    assert input.layer_type != LayerType.DATA
    assert isinstance(input.activation, SoftmaxActivation)
    assert label.layer_type == LayerType.DATA
4040 4041 4042

    ipts, parents = __cost_input__(input, label, weight)

Q
qijun 已提交
4043 4044 4045 4046
    Layer(
        name=name,
        type="multi-class-cross-entropy",
        inputs=ipts,
4047
        coeff=coeff,
Q
qijun 已提交
4048
        **ExtraLayerAttribute.to_kwargs(layer_attr))
Z
zhangjinchao01 已提交
4049 4050 4051 4052 4053 4054 4055 4056 4057 4058

    def __add_evaluator__(e):
        assert callable(e)
        assert hasattr(e, 'is_evaluator')
        assert isinstance(e.is_evaluator, bool)
        assert e.is_evaluator
        assert hasattr(e, "for_classification")
        assert isinstance(e.for_classification, bool)
        assert e.for_classification

4059
        e(name=e.__name__, input=input, label=label, weight=weight)
Z
zhangjinchao01 已提交
4060

4061
    if not isinstance(evaluator, collections.Sequence):
Z
zhangjinchao01 已提交
4062 4063 4064 4065 4066
        evaluator = [evaluator]

    for each_evaluator in evaluator:
        __add_evaluator__(each_evaluator)

X
xuwei06 已提交
4067
    return LayerOutput(name, LayerType.COST, parents=parents, size=1)
Z
zhangjinchao01 已提交
4068

4069

Q
qijun 已提交
4070 4071 4072 4073 4074 4075 4076 4077 4078
def conv_operator(img,
                  filter,
                  filter_size,
                  num_filters,
                  num_channels=None,
                  stride=1,
                  padding=0,
                  filter_size_y=None,
                  stride_y=None,
4079 4080
                  padding_y=None,
                  trans=False):
Z
zhangjinchao01 已提交
4081 4082 4083 4084 4085 4086 4087 4088 4089 4090
    """
    Different from img_conv_layer, conv_op is an Operator, which can be used
    in mixed_layer. And conv_op takes two inputs to perform convolution.
    The first input is the image and the second is filter kernel. It only
    support GPU mode.

    The example usage is:

    .. code-block:: python

4091 4092
       op = conv_operator(img=input1,
                          filter=input2,
4093
                          filter_size=3,
Z
zhangjinchao01 已提交
4094 4095 4096
                          num_filters=64,
                          num_channels=64)

4097 4098 4099 4100
    :param img: input image
    :type img: LayerOutput
    :param filter: input filter
    :type filter: LayerOutput
Z
zhangjinchao01 已提交
4101 4102
    :param filter_size: The x dimension of a filter kernel.
    :type filter_size: int
C
caoying03 已提交
4103 4104 4105
    :param filter_size_y: The y dimension of a filter kernel. Since
                        PaddlePaddle now supports rectangular filters,
                        the filter's shape can be (filter_size, filter_size_y).
Z
zhangjinchao01 已提交
4106
    :type filter_size_y: int
4107 4108
    :param num_filters: channel of output data.
    :type num_filters: int
4109 4110
    :param num_channels: channel of input data.
    :type num_channels: int
Z
zhangjinchao01 已提交
4111
    :param stride: The x dimension of the stride.
L
luotao02 已提交
4112
    :type stride: int
Z
zhangjinchao01 已提交
4113
    :param stride_y: The y dimension of the stride.
L
luotao02 已提交
4114
    :type stride_y: int
Z
zhangjinchao01 已提交
4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127
    :param padding: The x dimension of padding.
    :type padding: int
    :param padding_y: The y dimension of padding.
    :type padding_y: int
    :return: A ConvOperator Object.
    :rtype: ConvOperator
    """
    if filter_size_y is None:
        filter_size_y = filter_size
    if stride_y is None:
        stride_y = stride
    if padding_y is None:
        padding_y = padding
4128

4129 4130
    if num_channels is None:
        num_channels = img.num_filters
4131 4132 4133

    assert isinstance(filter, LayerOutput)
    if filter.size is not None:
4134
        filter.size = filter_size * filter_size_y * num_filters * num_channels
4135

4136 4137 4138
    opCls = ConvTransOperator if trans else ConvOperator

    op = opCls(
Q
qijun 已提交
4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149
        input_layer_names=[img.name, filter.name],
        num_filters=num_filters,
        conv_conf=Conv(
            filter_size=filter_size,
            padding=padding,
            stride=stride,
            channels=num_channels,
            filter_size_y=filter_size_y,
            padding_y=padding_y,
            stride_y=stride_y,
            groups=1))
4150

4151
    op.origin = [img, filter]
Z
zhangjinchao01 已提交
4152 4153
    return op

Q
qijun 已提交
4154

4155
@wrap_param_attr_default()
Q
qijun 已提交
4156 4157 4158 4159 4160 4161 4162 4163 4164 4165
def conv_projection(input,
                    filter_size,
                    num_filters,
                    num_channels=None,
                    stride=1,
                    padding=0,
                    filter_size_y=None,
                    stride_y=None,
                    padding_y=None,
                    groups=1,
4166 4167
                    param_attr=None,
                    trans=False):
4168 4169 4170 4171 4172 4173 4174 4175 4176
    """
    Different from img_conv_layer and conv_op, conv_projection is an Projection,
    which can be used in mixed_layer and conat_layer. It use cudnn to implement
    conv and only support GPU mode.

    The example usage is:

    .. code-block:: python

D
dangqingqing 已提交
4177
       proj = conv_projection(input=input1,
4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191
                              filter_size=3,
                              num_filters=64,
                              num_channels=64)

    :param input: input layer
    :type input: LayerOutput
    :param filter_size: The x dimension of a filter kernel.
    :type filter_size: int
    :param filter_size_y: The y dimension of a filter kernel. Since
                          PaddlePaddle now supports rectangular filters,
                          the filter's shape can be (filter_size, filter_size_y).
    :type filter_size_y: int
    :param num_filters: channel of output data.
    :type num_filters: int
4192 4193
    :param num_channels: channel of input data.
    :type num_channels: int
4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205
    :param stride: The x dimension of the stride.
    :type stride: int
    :param stride_y: The y dimension of the stride.
    :type stride_y: int
    :param padding: The x dimension of padding.
    :type padding: int
    :param padding_y: The y dimension of padding.
    :type padding_y: int
    :param groups: The group number.
    :type groups: int
    :param param_attr: Convolution param attribute. None means default attribute
    :type param_attr: ParameterAttribute
4206 4207
    :param trans: whether it is convTrans or conv
    :type trans: boolean
4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237
    :return: A DotMulProjection Object.
    :rtype: DotMulProjection
    """
    if num_channels is None:
        assert input.num_filters is not None
        num_channels = input.num_filters

    if filter_size_y is None:
        if isinstance(filter_size, collections.Sequence):
            assert len(filter_size) == 2
            filter_size, filter_size_y = filter_size
        else:
            filter_size_y = filter_size

    if stride_y is None:
        if isinstance(stride, collections.Sequence):
            assert len(stride) == 2
            stride, stride_y = stride
        else:
            stride_y = stride

    if padding_y is None:
        if isinstance(padding, collections.Sequence):
            assert len(padding) == 2
            padding, padding_y = padding
        else:
            padding_y = padding

    if param_attr.attr.get('initial_smart'):
        # special initial for conv layers.
Q
qijun 已提交
4238
        init_w = (2.0 / (filter_size**2 * num_channels))**0.5
4239 4240 4241 4242 4243
        param_attr.attr["initial_mean"] = 0.0
        param_attr.attr["initial_std"] = init_w
        param_attr.attr["initial_strategy"] = 0
        param_attr.attr["initial_smart"] = False

4244 4245 4246
    projCls = ConvTransProjection if trans else ConvProjection

    proj = projCls(
Q
qijun 已提交
4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258
        input_layer_name=input.name,
        num_filters=num_filters,
        conv_conf=Conv(
            filter_size=filter_size,
            padding=padding,
            stride=stride,
            channels=num_channels,
            filter_size_y=filter_size_y,
            padding_y=padding_y,
            stride_y=stride_y,
            groups=groups),
        **param_attr.attr)
4259 4260 4261 4262

    proj.origin = input
    return proj

Z
zhangjinchao01 已提交
4263

D
dangqingqing 已提交
4264 4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277 4278 4279 4280
@wrap_name_default("pad")
@layer_support()
def pad_layer(input,
              pad_c=None,
              pad_h=None,
              pad_w=None,
              name=None,
              layer_attr=None):
    """
    This operation pads zeros to the input data according to pad_c,pad_h
    and pad_w. pad_c, pad_h, pad_w specifies the which dimension and size
    of padding. And the input data shape is NCHW.

    For example, pad_c=[2,3] means padding 2 zeros before the
    input data and 3 zeros after the input data in channel dimension.
    pad_h means padding zeros in height dimension. pad_w means padding zeros
    in width dimension.
4281

D
dangqingqing 已提交
4282
    For example,
4283

4284 4285 4286 4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304
    .. code-block:: python

       input(2,2,2,3)  = [
                           [ [[1,2,3], [3,4,5]],
                             [[2,3,5], [1,6,7]] ],
                           [ [[4,3,1], [1,8,7]],
                             [[3,8,9], [2,3,5]] ]
                         ]

       pad_c=[1,1], pad_h=[0,0], pad_w=[0,0]

       output(2,4,2,3) = [
                           [ [[0,0,0], [0,0,0]],
                             [[1,2,3], [3,4,5]],
                             [[2,3,5], [1,6,7]],
                             [[0,0,0], [0,0,0]] ],
                           [ [[0,0,0], [0,0,0]],
                             [[4,3,1], [1,8,7]],
                             [[3,8,9], [2,3,5]],
                             [[0,0,0], [0,0,0]] ]
                         ]
D
dangqingqing 已提交
4305 4306

    The simply usage is:
D
dangqingqing 已提交
4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367

    .. code-block:: python

       pad = pad_layer(input=ipt,
                       pad_c=[4,4],
                       pad_h=[0,0],
                       pad_w=[2,2])

    :param input: layer's input.
    :type input: LayerOutput
    :param pad_c: padding size in channel dimension.
    :type pad_c: list|None
    :param pad_h: padding size in height dimension.
    :type pad_h: list|None
    :param pad_w: padding size in width dimension.
    :type pad_w: list|None
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
    :param name: layer name.
    :type name: basestring
    :return: LayerOutput object.
    :rtype: LayerOutput
    """
    if pad_c is not None:
        assert isinstance(pad_c, collections.Sequence) and len(pad_c) == 2
    else:
        pad_c = [0, 0]

    if pad_h is not None:
        assert isinstance(pad_h, collections.Sequence) and len(pad_h) == 2
    else:
        pad_h = [0, 0]

    if pad_w is not None:
        assert isinstance(pad_w, collections.Sequence) and len(pad_w) == 2
    else:
        pad_w = [0, 0]

    assert input.num_filters is not None
    in_ch = input.num_filters
    out_ch = in_ch + pad_c[0] + pad_c[1]

    l = Layer(
        name=name,
        type=LayerType.PAD_LAYER,
        inputs=Input(
            input.name,
            pad=Pad(
                channels=in_ch,
                pad_c=pad_c,
                pad_h=pad_h,
                pad_w=pad_w, )),
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name,
        layer_type=LayerType.PAD_LAYER,
        parents=[input],
        num_filters=out_ch,
        size=l.config.size)


Z
zhangjinchao01 已提交
4368
@wrap_name_default()
L
luotao1 已提交
4369 4370
@layer_support()
def conv_shift_layer(a, b, name=None, layer_attr=None):
Z
zhangjinchao01 已提交
4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381
    """
    This layer performs cyclic convolution for two input. For example:
      - a[in]: contains M elements.
      - b[in]: contains N elements (N should be odd).
      - c[out]: contains M elements.

    .. math::

        c[i] = \sum_{j=-(N-1)/2}^{(N-1)/2}a_{i+j} * b_{j}

    In this formular:
4382 4383 4384 4385
     - a's index is computed modulo M. When it is negative, then get item from
       the right side (which is the end of array) to the left.
     - b's index is computed modulo N. When it is negative, then get item from
       the right size (which is the end of array) to the left.
Z
zhangjinchao01 已提交
4386 4387 4388 4389 4390

    The example usage is:

    .. code-block:: python

L
Luo Tao 已提交
4391
       conv_shift = conv_shift_layer(a=layer1, b=layer2)
Z
zhangjinchao01 已提交
4392 4393 4394

    :param name: layer name
    :type name: basestring
4395 4396
    :param a: Input layer a.
    :type a: LayerOutput
L
Luo Tao 已提交
4397
    :param b: input layer b.
4398
    :type b: LayerOutput
L
luotao1 已提交
4399 4400
    :param layer_attr: layer's extra attribute.
    :type layer_attr: ExtraLayerAttribute
D
dangqingqing 已提交
4401
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
4402 4403
    :rtype: LayerOutput
    """
4404 4405
    assert isinstance(a, LayerOutput) and isinstance(b, LayerOutput)
    assert b.size is None or b.size % 2 == 1  # size of b must be odd.
Z
zhangjinchao01 已提交
4406 4407 4408
    Layer(
        name=name,
        type=LayerType.CONV_SHIFT_LAYER,
4409
        inputs=[a.name, b.name],
Q
qijun 已提交
4410
        **ExtraLayerAttribute.to_kwargs(layer_attr))
Z
zhangjinchao01 已提交
4411

Q
qijun 已提交
4412 4413
    return LayerOutput(
        name, LayerType.CONV_SHIFT_LAYER, parents=[a, b], size=a.size)
Z
zhangjinchao01 已提交
4414 4415 4416 4417 4418


@wrap_name_default()
@wrap_param_attr_default()
@wrap_bias_attr_default()
4419
@wrap_act_default(act=LinearActivation())
Z
zhangjinchao01 已提交
4420
@layer_support(ERROR_CLIPPING, DROPOUT)
Q
qijun 已提交
4421 4422 4423 4424 4425 4426 4427 4428
def tensor_layer(a,
                 b,
                 size,
                 act=None,
                 name=None,
                 param_attr=None,
                 bias_attr=None,
                 layer_attr=None):
Z
zhangjinchao01 已提交
4429 4430 4431 4432 4433
    """
    This layer performs tensor operation for two input.
    For example, each sample:

    .. math::
4434
       y_{i} = a * W_{i} * {b^\mathrm{T}}, i=0,1,...,K-1
Z
zhangjinchao01 已提交
4435 4436

    In this formular:
4437 4438
      - :math:`a`: the first input contains M elements.
      - :math:`b`: the second input contains N elements.
Z
zhangjinchao01 已提交
4439 4440
      - :math:`y_{i}`: the i-th element of y.
      - :math:`W_{i}`: the i-th learned weight, shape if [M, N]
4441
      - :math:`b^\mathrm{T}`: the transpose of :math:`b_{2}`.
Z
zhangjinchao01 已提交
4442 4443 4444 4445 4446

    The simple usage is:

    .. code-block:: python

4447
       tensor = tensor_layer(a=layer1, b=layer2, size=1000)
Z
zhangjinchao01 已提交
4448 4449 4450

    :param name: layer name
    :type name: basestring
4451 4452 4453 4454
    :param a: Input layer a.
    :type a: LayerOutput
    :param b: input layer b.
    :type b: LayerOutput
Z
zhangjinchao01 已提交
4455
    :param size: the layer dimension.
L
luotao02 已提交
4456
    :type size: int.
Z
zhangjinchao01 已提交
4457 4458 4459
    :param act: Activation Type. Default is tanh.
    :type act: BaseActivation
    :param param_attr: The Parameter Attribute.
4460
    :type param_attr: ParameterAttribute
Z
zhangjinchao01 已提交
4461 4462 4463 4464 4465 4466
    :param bias_attr: The Bias Attribute. If no bias, then pass False or
                      something not type of ParameterAttribute. None will get a
                      default Bias.
    :type bias_attr: ParameterAttribute|None|Any
    :param layer_attr: Extra Layer config.
    :type layer_attr: ExtraLayerAttribute|None
D
dangqingqing 已提交
4467
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
4468 4469
    :rtype: LayerOutput
    """
4470
    assert isinstance(a, LayerOutput) and isinstance(b, LayerOutput)
Z
zhangjinchao01 已提交
4471 4472 4473 4474 4475 4476
    Layer(
        name=name,
        size=size,
        type=LayerType.TENSOR_LAYER,
        active_type=act.name,
        bias=ParamAttr.to_bias(bias_attr),
Q
qijun 已提交
4477 4478 4479 4480
        inputs=[Input(a.name, **param_attr.attr), Input(b.name)],
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name, LayerType.TENSOR_LAYER, parents=[a, b], activation=act, size=size)
Z
zhangjinchao01 已提交
4481 4482 4483 4484 4485 4486


@wrap_name_default()
@wrap_param_attr_default()
@wrap_bias_attr_default()
@wrap_act_default()
L
luotao1 已提交
4487
@layer_support()
Q
qijun 已提交
4488 4489
def selective_fc_layer(input,
                       size,
L
Luo Tao 已提交
4490
                       select=None,
Q
qijun 已提交
4491 4492
                       act=None,
                       name=None,
Z
zhangjinchao01 已提交
4493 4494 4495
                       pass_generation=False,
                       has_selected_colums=True,
                       mul_ratio=0.02,
Q
qijun 已提交
4496 4497 4498
                       param_attr=None,
                       bias_attr=None,
                       layer_attr=None):
Z
zhangjinchao01 已提交
4499 4500 4501 4502 4503 4504 4505 4506 4507 4508
    """
    Selectived fully connected layer. Different from fc_layer, the output
    of this layer maybe sparse. It requires an additional input to indicate
    several selected columns for output. If the selected columns is not
    specified, selective_fc_layer acts exactly like fc_layer.

    The simple usage is:

    .. code-block:: python

4509
       sel_fc = selective_fc_layer(input=input, size=128, act=TanhActivation())
Z
zhangjinchao01 已提交
4510 4511 4512 4513 4514

    :param name: The Layer Name.
    :type name: basestring
    :param input: The input layer.
    :type input: LayerOutput|list|tuple
4515 4516
    :param select: The select layer. The output of select layer should be a
                   sparse binary matrix, and treat as the mask of selective fc.
L
Luo Tao 已提交
4517
                   If is None, acts exactly like fc_layer.
4518
    :type select: LayerOutput
Z
zhangjinchao01 已提交
4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530
    :param size: The layer dimension.
    :type size: int
    :param act: Activation Type. Default is tanh.
    :type act: BaseActivation
    :param param_attr: The Parameter Attribute.
    :type param_attr: ParameterAttribute
    :param bias_attr: The Bias Attribute. If no bias, then pass False or
                      something not type of ParameterAttribute. None will get a
                      default Bias.
    :type bias_attr: ParameterAttribute|None|Any
    :param layer_attr: Extra Layer config.
    :type layer_attr: ExtraLayerAttribute|None
D
dangqingqing 已提交
4531
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
4532 4533 4534 4535
    :rtype: LayerOutput
    """
    if isinstance(input, LayerOutput):
        input = [input]
4536
        assert not isinstance(param_attr, collections.Sequence)
Z
zhangjinchao01 已提交
4537 4538
        param_attr = [param_attr]
    else:
4539
        if isinstance(param_attr, collections.Sequence):
Z
zhangjinchao01 已提交
4540 4541 4542 4543
            assert len(input) == len(param_attr)
        else:
            param_attr = [copy.deepcopy(param_attr) for _ in range(len(input))]

4544 4545 4546 4547
    assert isinstance(input, collections.Sequence)
    assert isinstance(select, LayerOutput)
    if select.size is not None:
        assert select.size == size
Z
zhangjinchao01 已提交
4548
    Layer(
Q
qijun 已提交
4549 4550 4551
        inputs=[
            Input(ipt.name, **attr.attr) for ipt, attr in zip(input, param_attr)
        ] + [select.name],
Z
zhangjinchao01 已提交
4552 4553 4554
        name=name,
        type=LayerType.SEL_FC_LAYER,
        size=size,
4555
        bias=ParameterAttribute.to_bias(bias_attr),
Z
zhangjinchao01 已提交
4556 4557 4558 4559
        active_type=act.name,
        selective_fc_pass_generation=pass_generation,
        has_selected_colums=has_selected_colums,
        selective_fc_full_mul_ratio=mul_ratio,
Q
qijun 已提交
4560 4561 4562 4563 4564 4565 4566
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name,
        LayerType.SEL_FC_LAYER,
        list(input) + [select],
        activation=act,
        size=size)
Z
zhangjinchao01 已提交
4567 4568 4569


@wrap_name_default()
L
luotao1 已提交
4570 4571
@layer_support()
def sampling_id_layer(input, name=None, layer_attr=None):
Z
zhangjinchao01 已提交
4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585
    """
    A layer for sampling id from multinomial distribution from the input layer.
    Sampling one id for one sample.

    The simple usage is:

    .. code-block:: python

       samping_id = sampling_id_layer(input=input)

    :param input: The input layer.
    :type input: LayerOutput
    :param name: The Layer Name.
    :type name: basestring
L
luotao1 已提交
4586 4587
    :param layer_attr: Extra Layer config.
    :type layer_attr: ExtraLayerAttribute|None
D
dangqingqing 已提交
4588
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
4589 4590
    :rtype: LayerOutput
    """
X
xuwei06 已提交
4591
    l = Layer(
Z
zhangjinchao01 已提交
4592 4593 4594
        name=name,
        type=LayerType.SAMPLING_ID_LAYER,
        inputs=[Input(input.name)],
Q
qijun 已提交
4595 4596 4597
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name, LayerType.SAMPLING_ID_LAYER, input, size=l.config.size)
Z
zhangjinchao01 已提交
4598 4599 4600


@wrap_name_default()
L
luotao1 已提交
4601
@layer_support()
Q
qijun 已提交
4602 4603 4604 4605
def slope_intercept_layer(input,
                          name=None,
                          slope=1.0,
                          intercept=0.0,
L
luotao1 已提交
4606
                          layer_attr=None):
Z
zhangjinchao01 已提交
4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627
    """
    This layer for applying a slope and an intercept to the input
    element-wise. There is no activation and weight.

    ..  math::
        y = slope * x + intercept

    The simple usage is:

    .. code-block:: python

       scale = slope_intercept_layer(input=input, slope=-1.0, intercept=1.0)

    :param input: The input layer.
    :type input: LayerOutput
    :param name: The Layer Name.
    :type name: basestring
    :param slope: the scale factor.
    :type slope: float.
    :param intercept: the offset.
    :type intercept: float.
L
luotao1 已提交
4628 4629
    :param layer_attr: Extra Layer config.
    :type layer_attr: ExtraLayerAttribute|None
D
dangqingqing 已提交
4630
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
4631 4632 4633 4634 4635 4636 4637 4638
    :rtype: LayerOutput
    """
    Layer(
        name=name,
        type=LayerType.SLOPE_INTERCEPT_LAYER,
        slope=slope,
        intercept=intercept,
        inputs=[Input(input.name)],
Q
qijun 已提交
4639 4640 4641
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name, LayerType.SLOPE_INTERCEPT_LAYER, input, size=input.size)
Z
zhangjinchao01 已提交
4642 4643 4644


@wrap_name_default()
L
luotao1 已提交
4645
@layer_support()
Q
qijun 已提交
4646
def linear_comb_layer(weights, vectors, size=None, name=None, layer_attr=None):
Z
zhangjinchao01 已提交
4647
    """
4648 4649 4650 4651
    A layer for weighted sum of vectors takes two inputs.
      - Input: size of weights is M
               size of vectors is M*N
      - Output: a vector of size=N
Z
zhangjinchao01 已提交
4652 4653 4654

    .. math::

4655
       z(i) = \sum_{j=0}^{M-1} x(j) y(i+Nj)
4656

4657 4658 4659 4660 4661
    where :math:`0 \le i \le N-1`

    Or in the matrix notation:

    .. math::
Z
zhangjinchao01 已提交
4662

4663
       z = x^\mathrm{T} Y
Z
zhangjinchao01 已提交
4664 4665

    In this formular:
4666 4667 4668 4669 4670 4671
      - :math:`x`: weights
      - :math:`y`: vectors.
      - :math:`z`: the output.

    Note that the above computation is for one sample. Multiple samples are
    processed in one batch.
Z
zhangjinchao01 已提交
4672 4673 4674 4675 4676

    The simple usage is:

    .. code-block:: python

4677
       linear_comb = linear_comb_layer(weights=weight, vectors=vectors,
Z
zhangjinchao01 已提交
4678 4679
                                       size=elem_dim)

4680 4681 4682 4683
    :param weights: The weight layer.
    :type weights: LayerOutput
    :param vectors: The vector layer.
    :type vectors: LayerOutput
Z
zhangjinchao01 已提交
4684 4685 4686 4687
    :param size: the dimension of this layer.
    :type size: int
    :param name: The Layer Name.
    :type name: basestring
L
luotao1 已提交
4688 4689
    :param layer_attr: Extra Layer config.
    :type layer_attr: ExtraLayerAttribute|None
D
dangqingqing 已提交
4690
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
4691 4692
    :rtype: LayerOutput
    """
4693 4694 4695 4696
    assert isinstance(weights, LayerOutput) and isinstance(vectors, LayerOutput)
    if vectors.size is not None and weights.size is not None:
        assert vectors.size % weights.size == 0
        if size is None:
Q
qijun 已提交
4697
            size = vectors.size / weights.size
4698 4699
        else:
            assert size == vectors.size / weights.size
Z
zhangjinchao01 已提交
4700 4701
    Layer(
        name=name,
4702
        type=LayerType.LINEAR_COMBINATION_LAYER,
Z
zhangjinchao01 已提交
4703
        size=size,
4704
        inputs=[Input(weights.name), Input(vectors.name)],
Q
qijun 已提交
4705 4706 4707
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name, LayerType.LINEAR_COMBINATION_LAYER, [weights, vectors], size=size)
4708

4709

4710
convex_comb_layer = linear_comb_layer
Z
zhangjinchao01 已提交
4711

4712

Z
zhangjinchao01 已提交
4713
@wrap_name_default()
L
luotao1 已提交
4714
@layer_support()
Z
zhangjinchao01 已提交
4715 4716 4717 4718 4719 4720 4721
def block_expand_layer(input,
                       block_x=0,
                       block_y=0,
                       stride_x=0,
                       stride_y=0,
                       padding_x=0,
                       padding_y=0,
4722
                       num_channels=None,
L
luotao1 已提交
4723 4724
                       name=None,
                       layer_attr=None):
Z
zhangjinchao01 已提交
4725 4726
    """
    Expand feature map to minibatch matrix.
4727
       - matrix width is: block_y * block_x * num_channels
L
luotao02 已提交
4728
       - matirx height is: outputH * outputW
Z
zhangjinchao01 已提交
4729 4730 4731 4732 4733 4734 4735 4736 4737 4738

    .. math::

       outputH = 1 + (2 * padding_y + imgSizeH - block_y + stride_y - 1) / stride_y

       outputW = 1 + (2 * padding_x + imgSizeW - block_x + stride_x - 1) / stride_x

    The expand method is the same with ExpandConvLayer, but saved the transposed
    value. After expanding, output.sequenceStartPositions will store timeline.
    The number of time steps are outputH * outputW and the dimension of each
4739
    time step is block_y * block_x * num_channels. This layer can be used after
Z
zhangjinchao01 已提交
4740 4741
    convolution neural network, and before recurrent neural network.

4742 4743 4744 4745
    The simple usage is:

    .. code-block:: python

L
Luo Tao 已提交
4746
       block_expand = block_expand_layer(input=layer,
4747
                                         num_channels=128,
4748 4749 4750 4751 4752
                                         stride_x=1,
                                         stride_y=1,
                                         block_x=1,
                                         block_x=3)

Z
zhangjinchao01 已提交
4753 4754
    :param input: The input layer.
    :type input: LayerOutput
4755 4756
    :param num_channels: The channel number of input layer.
    :type num_channels: int|None
Z
zhangjinchao01 已提交
4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770
    :param block_x: The width of sub block.
    :type block_x: int
    :param block_y: The width of sub block.
    :type block_y: int
    :param stride_x: The stride size in horizontal direction.
    :type stride_x: int
    :param stride_y: The stride size in vertical direction.
    :type stride_y: int
    :param padding_x: The padding size in horizontal direction.
    :type padding_x: int
    :param padding_y: The padding size in vertical direction.
    :type padding_y: int
    :param name: The name of this layer, which can not specify.
    :type name: None|basestring.
L
luotao1 已提交
4771 4772
    :param layer_attr: Extra Layer config.
    :type layer_attr: ExtraLayerAttribute|None
D
dangqingqing 已提交
4773
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
4774 4775
    :rtype: LayerOutput
    """
4776 4777 4778
    if num_channels is None:
        assert input.num_filters is not None
        num_channels = input.num_filters
Q
qijun 已提交
4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795
    l = Layer(
        name=name,
        inputs=Input(
            input.name,
            block_expand=BlockExpand(
                channels=num_channels,
                block_x=block_x,
                block_y=block_y,
                stride_x=stride_x,
                stride_y=stride_y,
                padding_x=padding_x,
                padding_y=padding_y)),
        type=LayerType.BLOCK_EXPAND,
        **ExtraLayerAttribute.to_kwargs(layer_attr))

    return LayerOutput(
        name, LayerType.BLOCK_EXPAND, parents=[input], size=l.config.size)
Z
zhangjinchao01 已提交
4796 4797


4798 4799
@wrap_name_default()
@layer_support()
4800
def maxout_layer(input, groups, num_channels=None, name=None, layer_attr=None):
4801 4802 4803 4804 4805
    """
    A layer to do max out on conv layer output.
      - Input: output of a conv layer.
      - Output: feature map size same as input. Channel is (input channel) / groups.

4806
    So groups should be larger than 1, and the num of channels should be able
4807 4808
    to devided by groups.

X
xuwei06 已提交
4809 4810 4811 4812 4813 4814 4815 4816
    .. math::
       y_{si+j} = \max_k x_{gsi + sk + j}
       g = groups
       s = input.size / num_channels
       0 \le i < num_channels / groups
       0 \le j < s
       0 \le k < groups

4817
    Please refer to Paper:
4818 4819 4820 4821
      - Maxout Networks: http://www.jmlr.org/proceedings/papers/v28/goodfellow13.pdf
      - Multi-digit Number Recognition from Street View \
        Imagery using Deep Convolutional Neural Networks: \
        https://arxiv.org/pdf/1312.6082v4.pdf
4822

4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851
    The simple usage is:

    .. code-block:: python

       maxout = maxout_layer(input,
                             num_channels=128,
                             groups=4)

    :param input: The input layer.
    :type input: LayerOutput
    :param num_channels: The channel number of input layer. If None will be set
                     automatically from previous output.
    :type num_channels: int|None
    :param groups: The group number of input layer.
    :type groups: int
    :param name: The name of this layer, which can not specify.
    :type name: None|basestring.
    :param layer_attr: Extra Layer attribute.
    :type layer_attr: ExtraLayerAttribute
    :return: LayerOutput object.
    :rtype: LayerOutput
    """
    assert input.layer_type == LayerType.CONV_LAYER
    assert isinstance(input.activation, LinearActivation)
    assert groups > 1
    if num_channels is None:
        assert input.num_filters is not None
        num_channels = input.num_filters
    assert num_channels % groups == 0
Q
qijun 已提交
4852 4853 4854 4855 4856 4857 4858 4859 4860
    l = Layer(
        name=name,
        inputs=Input(
            input.name, maxout=MaxOut(
                channels=num_channels, groups=groups)),
        type=LayerType.MAXOUT,
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name, LayerType.MAXOUT, parents=[input], size=l.config.size)
4861 4862


Z
zhangjinchao01 已提交
4863
@wrap_name_default()
L
luotao1 已提交
4864
@layer_support()
Q
qijun 已提交
4865 4866 4867 4868 4869
def ctc_layer(input,
              label,
              size=None,
              name=None,
              norm_by_times=False,
L
luotao1 已提交
4870
              layer_attr=None):
Z
zhangjinchao01 已提交
4871 4872 4873 4874 4875
    """
    Connectionist Temporal Classification (CTC) is designed for temporal
    classication task. That is, for sequence labeling problems where the
    alignment between the inputs and the target labels is unknown.

4876 4877
    More details can be found by referring to `Connectionist Temporal
    Classification: Labelling Unsegmented Sequence Data with Recurrent
4878 4879
    Neural Networks <http://machinelearning.wustl.edu/mlpapers/paper_files/
    icml2006_GravesFGS06.pdf>`_
4880 4881 4882 4883 4884 4885 4886 4887

    Note:
        Considering the 'blank' label needed by CTC, you need to use
        (num_classes + 1) as the input size. num_classes is the category number.
        And the 'blank' is the last category index. So the size of 'input' layer, such as
        fc_layer with softmax activation, should be num_classes + 1. The size of ctc_layer
        should also be num_classes + 1.

C
caoying03 已提交
4888
    The example usage is:
Z
zhangjinchao01 已提交
4889 4890 4891 4892 4893 4894 4895 4896

    .. code-block:: python

      ctc = ctc_layer(input=input,
                      label=label,
                      size=9055,
                      norm_by_times=True)

4897
    :param input: The input layer.
Z
zhangjinchao01 已提交
4898 4899 4900
    :type input: LayerOutput
    :param label: The data layer of label with variable length.
    :type label: LayerOutput
4901
    :param size: category numbers + 1.
Z
zhangjinchao01 已提交
4902
    :type size: int
4903 4904
    :param name: The name of this layer
    :type name: basestring|None
Z
zhangjinchao01 已提交
4905 4906
    :param norm_by_times: Whether to normalization by times. False by default.
    :type norm_by_times: bool
L
luotao1 已提交
4907 4908
    :param layer_attr: Extra Layer config.
    :type layer_attr: ExtraLayerAttribute|None
D
dangqingqing 已提交
4909
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
4910 4911 4912 4913
    :rtype: LayerOutput
    """
    assert isinstance(input, LayerOutput)
    assert isinstance(label, LayerOutput)
4914 4915 4916 4917 4918
    if label.size is not None:
        if size is not None:
            assert size == label.size + 1
        else:
            size = label.size + 1
Z
zhangjinchao01 已提交
4919
    Layer(
4920 4921 4922 4923
        name=name,
        type=LayerType.CTC_LAYER,
        size=size,
        norm_by_times=norm_by_times,
L
luotao1 已提交
4924
        inputs=[input.name, label.name],
Q
qijun 已提交
4925
        **ExtraLayerAttribute.to_kwargs(layer_attr))
Z
zhangjinchao01 已提交
4926 4927
    return LayerOutput(name, LayerType.CTC_LAYER, [input, label], size=size)

4928

4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939
@wrap_name_default()
@layer_support()
def warp_ctc_layer(input,
                   label,
                   size=None,
                   name=None,
                   blank=0,
                   norm_by_times=False,
                   layer_attr=None):
    """
    A layer intergrating the open-source `warp-ctc
L
Liu Yiqun 已提交
4940
    <https://github.com/baidu-research/warp-ctc>`_ library, which is used in
4941
    `Deep Speech 2: End-toEnd Speech Recognition in English and Mandarin
L
Liu Yiqun 已提交
4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958
    <https://arxiv.org/pdf/1512.02595v1.pdf>`_, to compute Connectionist Temporal
    Classification (CTC) loss. Besides, another `warp-ctc
    <https://github.com/gangliao/warp-ctc>`_ repository, which is forked from
    the official one, is maintained to enable more compiling options. During the
    building process, PaddlePaddle will clone the source codes, build and
    install it to :code:`third_party/install/warpctc` directory.

    To use warp_ctc layer, you need to specify the path of :code:`libwarpctc.so`,
    using following methods:

    1. Set it in :code:`paddle.init` (python api) or :code:`paddle_init` (c api),
    such as :code:`paddle.init(use_gpu=True,
    warpctc_dir=your_paddle_source_dir/third_party/install/warpctc/lib)`.

    2. Set environment variable LD_LIBRARY_PATH on Linux or DYLD_LIBRARY_PATH
    on Mac OS. For instance, :code:`export
    LD_LIBRARY_PATH=your_paddle_source_dir/third_party/install/warpctc/lib:$LD_LIBRARY_PATH`.
4959 4960 4961 4962

    More details of CTC can be found by referring to `Connectionist Temporal
    Classification: Labelling Unsegmented Sequence Data with Recurrent
    Neural Networks <http://machinelearning.wustl.edu/mlpapers/paper_files/
L
Liu Yiqun 已提交
4963
    icml2006_GravesFGS06.pdf>`_.
4964 4965 4966

    Note:
        - Let num_classes represent the category number. Considering the 'blank'
L
Liu Yiqun 已提交
4967 4968 4969
          label needed by CTC, you need to use (num_classes + 1) as the input size.
          Thus, the size of both warp_ctc layer and 'input' layer should be set to
          num_classes + 1.
4970 4971
        - You can set 'blank' to any value ranged in [0, num_classes], which
          should be consistent as that used in your labels.
4972
        - As a native 'softmax' activation is interated to the warp-ctc library,
L
Luo Tao 已提交
4973
          'linear' activation is expected instead in the 'input' layer.
4974

C
caoying03 已提交
4975
    The example usage is:
4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019 5020

    .. code-block:: python

      ctc = warp_ctc_layer(input=input,
                           label=label,
                           size=1001,
                           blank=1000,
                           norm_by_times=False)

    :param input: The input layer.
    :type input: LayerOutput
    :param label: The data layer of label with variable length.
    :type label: LayerOutput
    :param size: category numbers + 1.
    :type size: int
    :param name: The name of this layer, which can not specify.
    :type name: basestring|None
    :param blank: the 'blank' label used in ctc
    :type blank: int
    :param norm_by_times: Whether to normalization by times. False by default.
    :type norm_by_times: bool
    :param layer_attr: Extra Layer config.
    :type layer_attr: ExtraLayerAttribute|None
    :return: LayerOutput object.
    :rtype: LayerOutput
    """
    assert isinstance(input, LayerOutput)
    assert isinstance(label, LayerOutput)
    if label.size is not None:
        if size is not None:
            assert size == label.size + 1
        else:
            size = label.size + 1
    Layer(
        name=name,
        type=LayerType.WARP_CTC_LAYER,
        size=size,
        blank=blank,
        norm_by_times=norm_by_times,
        inputs=[input.name, label.name],
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name, LayerType.WARP_CTC_LAYER, parents=[input, label], size=size)


Z
zhangjinchao01 已提交
5021
@wrap_name_default()
5022
@wrap_param_attr_default()
L
luotao1 已提交
5023
@layer_support()
Q
qijun 已提交
5024 5025 5026 5027 5028 5029
def crf_layer(input,
              label,
              size=None,
              weight=None,
              param_attr=None,
              name=None,
5030
              coeff=1.0,
L
luotao1 已提交
5031
              layer_attr=None):
Z
zhangjinchao01 已提交
5032 5033 5034 5035
    """
    A layer for calculating the cost of sequential conditional random
    field model.

C
caoying03 已提交
5036
    The example usage is:
Z
zhangjinchao01 已提交
5037 5038 5039 5040 5041 5042 5043 5044 5045 5046

    .. code-block:: python

      crf = crf_layer(input=input,
                      label=label,
                      size=label_dim)

    :param input: The first input layer is the feature.
    :type input: LayerOutput
    :param label: The second input layer is label.
5047
    :type label: LayerOutput
Z
zhangjinchao01 已提交
5048 5049 5050 5051 5052 5053 5054 5055 5056
    :param size: The category number.
    :type size: int
    :param weight: The third layer is "weight" of each sample, which is an
                  optional argument.
    :type weight: LayerOutput
    :param param_attr: Parameter attribute. None means default attribute
    :type param_attr: ParameterAttribute
    :param name: The name of this layers. It is not necessary.
    :type name: None|basestring
5057 5058
    :param coeff: The coefficient affects the gradient in the backward.
    :type coeff: float
L
luotao1 已提交
5059 5060
    :param layer_attr: Extra Layer config.
    :type layer_attr: ExtraLayerAttribute|None
D
dangqingqing 已提交
5061
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
5062 5063 5064 5065 5066
    :rtype: LayerOutput
    """
    assert isinstance(input, LayerOutput)
    assert isinstance(label, LayerOutput)
    assert weight is None or isinstance(weight, LayerOutput)
5067 5068 5069 5070 5071 5072
    if input.size is not None and label.size is not None:
        assert input.size == label.size
        if size is None:
            size = input.size
        else:
            assert size == input.size
Z
zhangjinchao01 已提交
5073

Q
qijun 已提交
5074
    ipts = [Input(input.name, **param_attr.attr), Input(label.name)]
Z
zhangjinchao01 已提交
5075 5076 5077 5078
    if weight is not None:
        ipts.append(Input(weight.name))

    Layer(
5079 5080 5081 5082
        name=name,
        type=LayerType.CRF_LAYER,
        size=size,
        inputs=ipts,
5083
        coeff=coeff,
Q
qijun 已提交
5084
        **ExtraLayerAttribute.to_kwargs(layer_attr))
Z
zhangjinchao01 已提交
5085 5086 5087
    parents = [input, label]
    if weight is not None:
        parents.append(weight)
X
xuwei06 已提交
5088 5089 5090 5091
    # The size for LayerOutput means the dimension of the output.
    # It's different from the meaning of crf layer, which is the number of
    # classes.
    return LayerOutput(name, LayerType.CRF_LAYER, parents, size=1)
Z
zhangjinchao01 已提交
5092

5093

Z
zhangjinchao01 已提交
5094
@wrap_name_default()
5095
@wrap_param_attr_default()
L
luotao1 已提交
5096
@layer_support()
Q
qijun 已提交
5097 5098 5099 5100 5101
def crf_decoding_layer(input,
                       size,
                       label=None,
                       param_attr=None,
                       name=None,
L
luotao1 已提交
5102
                       layer_attr=None):
Z
zhangjinchao01 已提交
5103 5104 5105 5106 5107 5108 5109
    """
    A layer for calculating the decoding sequence of sequential conditional
    random field model. The decoding sequence is stored in output.ids.
    If a second input is provided, it is treated as the ground-truth label, and
    this layer will also calculate error. output.value[i] is 1 for incorrect
    decoding or 0 for correct decoding.

C
caoying03 已提交
5110
    The example usage is:
L
Luo Tao 已提交
5111 5112 5113 5114 5115 5116

    .. code-block:: python

      crf_decoding = crf_decoding_layer(input=input,
                                        size=label_dim)

Z
zhangjinchao01 已提交
5117 5118 5119 5120 5121 5122 5123 5124 5125 5126
    :param input: The first input layer.
    :type input: LayerOutput
    :param size: size of this layer.
    :type size: int
    :param label: None or ground-truth label.
    :type label: LayerOutput or None
    :param param_attr: Parameter attribute. None means default attribute
    :type param_attr: ParameterAttribute
    :param name: The name of this layers. It is not necessary.
    :type name: None|basestring
L
luotao1 已提交
5127 5128
    :param layer_attr: Extra Layer config.
    :type layer_attr: ExtraLayerAttribute|None
D
dangqingqing 已提交
5129
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
5130 5131 5132 5133 5134 5135
    :rtype: LayerOutput
    """

    assert isinstance(input, LayerOutput)
    assert label is None or isinstance(label, LayerOutput)

5136
    ipts = [Input(input.name, **param_attr.attr)]
Z
zhangjinchao01 已提交
5137 5138 5139 5140
    if label is not None:
        ipts.append(Input(label.name))

    Layer(
5141 5142 5143 5144
        name=name,
        type=LayerType.CRF_DECODING_LAYER,
        size=size,
        inputs=ipts,
Q
qijun 已提交
5145
        **ExtraLayerAttribute.to_kwargs(layer_attr))
Z
zhangjinchao01 已提交
5146 5147 5148
    parents = [input]
    if label is not None:
        parents.append(label)
X
xuwei06 已提交
5149 5150 5151 5152
    # The size for LayerOutput means the dimension of the output.
    # It's different from the meaning of crf layer, which is the number of
    # classes.
    return LayerOutput(name, LayerType.CRF_DECODING_LAYER, parents, size=1)
Z
zhangjinchao01 已提交
5153

Q
qijun 已提交
5154

Y
Yu Yang 已提交
5155
@wrap_act_default(act=SigmoidActivation())
5156
@wrap_bias_attr_default(has_bias=True)
5157
@wrap_param_attr_default()
5158 5159
@wrap_name_default()
@layer_support()
Q
qijun 已提交
5160 5161
def nce_layer(input,
              label,
C
caoying03 已提交
5162
              num_classes=None,
Y
Yu Yang 已提交
5163
              act=None,
5164
              param_attr=None,
Q
qijun 已提交
5165 5166 5167 5168 5169 5170
              weight=None,
              num_neg_samples=10,
              neg_distribution=None,
              name=None,
              bias_attr=None,
              layer_attr=None):
5171 5172 5173 5174 5175 5176 5177 5178 5179
    """
    Noise-contrastive estimation.
    Implements the method in the following paper:
    A fast and simple algorithm for training neural probabilistic language models.

    The example usage is:

    .. code-block:: python

C
caoying03 已提交
5180 5181
       cost = nce_layer(input=[layer1, layer2], label=layer2,
                        param_attr=[attr1, attr2], weight=layer3,
5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192
                        num_classes=3, neg_distribution=[0.1,0.3,0.6])

    :param name: layer name
    :type name: basestring
    :param input: input layers. It could be a LayerOutput of list/tuple of LayerOutput.
    :type input: LayerOutput|list|tuple|collections.Sequence
    :param label: label layer
    :type label: LayerOutput
    :param weight: weight layer, can be None(default)
    :type weight: LayerOutput
    :param num_classes: number of classes.
5193
    :type num_classes: int
Y
Yu Yang 已提交
5194 5195
    :param act: Activation, default is Sigmoid.
    :type act: BaseActivation
5196 5197
    :param param_attr: The Parameter Attribute|list.
    :type param_attr: ParameterAttribute
5198
    :param num_neg_samples: number of negative samples. Default is 10.
5199
    :type num_neg_samples: int
5200 5201 5202 5203 5204 5205 5206 5207 5208 5209 5210 5211 5212
    :param neg_distribution: The distribution for generating the random negative labels.
                             A uniform distribution will be used if not provided.
                             If not None, its length must be equal to num_classes.
    :type neg_distribution: list|tuple|collections.Sequence|None
    :param bias_attr: Bias parameter attribute. True if no bias.
    :type bias_attr: ParameterAttribute|None|False
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
    :return: layer name.
    :rtype: LayerOutput
    """
    if isinstance(input, LayerOutput):
        input = [input]
5213 5214 5215 5216 5217 5218 5219 5220
        assert not isinstance(param_attr, collections.Sequence)
        param_attr = [param_attr]
    else:
        if isinstance(param_attr, collections.Sequence):
            assert len(input) == len(param_attr)
        else:
            param_attr = [copy.deepcopy(param_attr) for _ in range(len(input))]

5221
    assert isinstance(input, collections.Sequence)
5222

5223 5224
    assert isinstance(label, LayerOutput)
    assert label.layer_type == LayerType.DATA
C
caoying03 已提交
5225 5226
    if num_classes is None:
        num_classes = label.size
5227 5228 5229
    if neg_distribution is not None:
        assert isinstance(neg_distribution, collections.Sequence)
        assert len(neg_distribution) == num_classes
5230
        assert abs(sum(neg_distribution) - 1.0) < 1e-5
Y
Yu Yang 已提交
5231 5232
    if not isinstance(act, BaseActivation):
        raise TypeError()
5233

5234 5235
    ipts_for_layer = []
    parents = []
5236
    for each_input, attr in zip(input, param_attr):
5237
        assert isinstance(each_input, LayerOutput)
5238
        ipts_for_layer.append(Input(each_input.name, **attr.attr))
5239 5240 5241 5242 5243 5244 5245 5246 5247 5248
        parents.append(each_input)
    ipts_for_layer.append(label.name)
    parents.append(label)

    if weight is not None:
        assert isinstance(weight, LayerOutput)
        assert weight.layer_type == LayerType.DATA
        ipts_for_layer.append(weight.name)
        parents.append(weight)

X
xuwei06 已提交
5249
    l = Layer(
5250 5251 5252 5253
        name=name,
        type=LayerType.NCE_LAYER,
        num_classes=num_classes,
        neg_sampling_dist=neg_distribution,
Y
Yu Yang 已提交
5254
        active_type=act.name,
5255 5256 5257
        num_neg_samples=num_neg_samples,
        inputs=ipts_for_layer,
        bias=ParamAttr.to_bias(bias_attr),
Q
qijun 已提交
5258 5259
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
Y
Yu Yang 已提交
5260 5261 5262 5263 5264
        name,
        LayerType.NCE_LAYER,
        parents=parents,
        size=l.config.size,
        activation=act)
Q
qijun 已提交
5265

5266

Z
zhangjinchao01 已提交
5267 5268 5269
"""
following are cost Layers.
"""
5270 5271


Z
zhangjinchao01 已提交
5272
@wrap_name_default()
L
luotao1 已提交
5273
@layer_support()
Q
qijun 已提交
5274 5275 5276 5277 5278 5279 5280
def rank_cost(left,
              right,
              label,
              weight=None,
              name=None,
              coeff=1.0,
              layer_attr=None):
Z
zhangjinchao01 已提交
5281
    """
5282
    A cost Layer for learning to rank using gradient descent. Details can refer
5283 5284
    to `papers <http://research.microsoft.com/en-us/um/people/cburges/papers/
    ICML_ranking.pdf>`_.
Z
zhangjinchao01 已提交
5285 5286 5287 5288 5289
    This layer contains at least three inputs. The weight is an optional
    argument, which affects the cost.

    .. math::

L
luotao02 已提交
5290
       C_{i,j} & = -\\tilde{P_{ij}} * o_{i,j} + log(1 + e^{o_{i,j}})
Z
zhangjinchao01 已提交
5291

L
luotao02 已提交
5292
       o_{i,j} & =  o_i - o_j
Z
zhangjinchao01 已提交
5293

L
luotao02 已提交
5294
       \\tilde{P_{i,j}} & = \\{0, 0.5, 1\\} \ or \ \\{0, 1\\}
Z
zhangjinchao01 已提交
5295 5296 5297 5298 5299 5300 5301 5302

    In this formula:
      - :math:`C_{i,j}` is the cross entropy cost.
      - :math:`\\tilde{P_{i,j}}` is the label. 1 means positive order
        and 0 means reverse order.
      - :math:`o_i` and :math:`o_j`: the left output and right output.
        Their dimension is one.

C
caoying03 已提交
5303
    The example usage is:
Z
zhangjinchao01 已提交
5304 5305 5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321 5322 5323

    .. code-block:: python

      cost = rank_cost(left=out_left,
                       right=out_right,
                       label=label)

    :param left: The first input, the size of this layer is 1.
    :type left: LayerOutput
    :param right: The right input, the size of this layer is 1.
    :type right: LayerOutput
    :param label: Label is 1 or 0, means positive order and reverse order.
    :type label: LayerOutput
    :param weight: The weight affects the cost, namely the scale of cost.
                   It is an optional argument.
    :type weight: LayerOutput
    :param name: The name of this layers. It is not necessary.
    :type name: None|basestring
    :param coeff: The coefficient affects the gradient in the backward.
    :type coeff: float
L
luotao1 已提交
5324 5325
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
D
dangqingqing 已提交
5326
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338
    :rtype: LayerOutput
    """
    assert left.size == 1
    assert right.size == 1
    assert label.size == 1

    ipts = [left.name, right.name, label.name]
    parents = [left, right, label]
    if weight is not None:
        ipts.append(weight.name)
        parents.append(weight)

Q
qijun 已提交
5339 5340 5341 5342 5343 5344
    Layer(
        name=name,
        type=LayerType.RANK_COST,
        inputs=ipts,
        coeff=coeff,
        **ExtraLayerAttribute.to_kwargs(layer_attr))
Z
zhangjinchao01 已提交
5345

X
xuwei06 已提交
5346
    return LayerOutput(name, LayerType.RANK_COST, parents=parents, size=1)
Z
zhangjinchao01 已提交
5347

5348

Z
zhangjinchao01 已提交
5349
@wrap_name_default()
L
luotao1 已提交
5350
@layer_support()
Q
qijun 已提交
5351 5352 5353 5354 5355 5356
def lambda_cost(input,
                score,
                name,
                NDCG_num=5,
                max_sort_size=-1,
                layer_attr=None):
Z
zhangjinchao01 已提交
5357 5358 5359
    """
    lambdaCost for lambdaRank LTR approach.

C
caoying03 已提交
5360
    The example usage is:
Z
zhangjinchao01 已提交
5361 5362 5363 5364 5365 5366 5367 5368

    .. code-block:: python

      cost = lambda_cost(input=input,
                         score=score,
                         NDCG_num=8,
                         max_sort_size=-1)

5369
    :param input: Samples of the same query should be loaded as sequence.
Z
zhangjinchao01 已提交
5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380
    :type input: LayerOutput
    :param score: The 2nd input. Score of each sample.
    :type input: LayerOutput
    :param NDCG_num: The size of NDCG (Normalized Discounted Cumulative Gain),
                     e.g., 5 for NDCG@5. It must be less than for equal to the
                     minimum size of lists.
    :type NDCG_num: int
    :param max_sort_size: The size of partial sorting in calculating gradient.
                          If max_sort_size = -1, then for each list, the
                          algorithm will sort the entire list to get gradient.
                          In other cases, max_sort_size must be greater than or
C
caoying03 已提交
5381 5382 5383
                          equal to NDCG_num. And if max_sort_size is greater
                          than the size of a list, the algorithm will sort the
                          entire list of get gradient.
Z
zhangjinchao01 已提交
5384 5385 5386
    :type max_sort_size: int
    :param name: The name of this layers. It is not necessary.
    :type name: None|basestring
L
luotao1 已提交
5387 5388
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
D
dangqingqing 已提交
5389
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
5390 5391
    :rtype: LayerOutput
    """
5392 5393 5394
    assert isinstance(input, LayerOutput) and isinstance(score, LayerOutput)
    if score.size is not None:
        assert score.size == 1
Q
qijun 已提交
5395 5396 5397 5398 5399 5400 5401
    Layer(
        name=name,
        type=LayerType.LAMBDA_COST,
        inputs=[input.name, score.name],
        NDCG_num=NDCG_num,
        max_sort_size=max_sort_size,
        **ExtraLayerAttribute.to_kwargs(layer_attr))
Z
zhangjinchao01 已提交
5402

Q
qijun 已提交
5403 5404
    return LayerOutput(
        name, LayerType.LAMBDA_COST, parents=[input, score], size=1)
Z
zhangjinchao01 已提交
5405

5406

Z
zhangjinchao01 已提交
5407
@wrap_name_default()
L
luotao1 已提交
5408
@layer_support()
5409 5410 5411 5412 5413 5414
def cross_entropy(input,
                  label,
                  name=None,
                  coeff=1.0,
                  weight=None,
                  layer_attr=None):
Z
zhangjinchao01 已提交
5415 5416 5417
    """
    A loss layer for multi class entropy.

C
caoying03 已提交
5418 5419
    The example usage is:

Z
zhangjinchao01 已提交
5420 5421
    .. code-block:: python

X
xuwei06 已提交
5422
       cost = cross_entropy(input=input_layer,
L
Luo Tao 已提交
5423
                            label=label_layer)
Z
zhangjinchao01 已提交
5424 5425 5426 5427 5428 5429 5430

    :param input: The first input layer.
    :type input: LayerOutput.
    :param label: The input label.
    :type input: LayerOutput.
    :param name: The name of this layers. It is not necessary.
    :type name: None|basestring.
5431 5432
    :param coeff: The cost is multiplied with coeff.
                  The coefficient affects the gradient in the backward.
Z
zhangjinchao01 已提交
5433
    :type coeff: float.
5434 5435 5436 5437
    :param weight: The cost of each sample is multiplied with each weight.
                   The weight should be a layer with size=1. Note that gradient
                   will not be calculated for weight.
    :type weight: LayerOutout
L
luotao1 已提交
5438 5439
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
D
dangqingqing 已提交
5440
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
5441 5442 5443
    :rtype: LayerOutput.
    """

5444
    ipts, parents = __cost_input__(input, label, weight)
Q
qijun 已提交
5445 5446 5447
    Layer(
        name=name,
        type=LayerType.CROSS_ENTROPY,
5448
        inputs=ipts,
Q
qijun 已提交
5449 5450
        coeff=coeff,
        **ExtraLayerAttribute.to_kwargs(layer_attr))
5451
    return LayerOutput(name, LayerType.CROSS_ENTROPY, parents=parents, size=1)
Z
zhangjinchao01 已提交
5452

5453

Z
zhangjinchao01 已提交
5454
@wrap_name_default()
L
luotao1 已提交
5455
@layer_support()
Q
qijun 已提交
5456 5457 5458 5459
def cross_entropy_with_selfnorm(input,
                                label,
                                name=None,
                                coeff=1.0,
L
luotao1 已提交
5460 5461
                                softmax_selfnorm_alpha=0.1,
                                layer_attr=None):
Z
zhangjinchao01 已提交
5462 5463
    """
    A loss layer for multi class entropy with selfnorm.
5464
    Input should be a vector of positive numbers, without normalization.
Z
zhangjinchao01 已提交
5465

C
caoying03 已提交
5466 5467
    The example usage is:

Z
zhangjinchao01 已提交
5468 5469
    .. code-block:: python

X
xuwei06 已提交
5470
       cost = cross_entropy_with_selfnorm(input=input_layer,
L
Luo Tao 已提交
5471
                                          label=label_layer)
Z
zhangjinchao01 已提交
5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482

    :param input: The first input layer.
    :type input: LayerOutput.
    :param label: The input label.
    :type input: LayerOutput.
    :param name: The name of this layers. It is not necessary.
    :type name: None|basestring.
    :param coeff: The coefficient affects the gradient in the backward.
    :type coeff: float.
    :param softmax_selfnorm_alpha: The scale factor affects the cost.
    :type softmax_selfnorm_alpha: float.
L
luotao1 已提交
5483 5484
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
D
dangqingqing 已提交
5485
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
5486 5487
    :rtype: LayerOutput.
    """
Q
qijun 已提交
5488 5489 5490 5491 5492 5493 5494
    Layer(
        name=name,
        type=LayerType.CROSS_ENTROPY_WITH_SELFNORM,
        inputs=[input.name, label.name],
        coeff=coeff,
        softmax_selfnorm_alpha=softmax_selfnorm_alpha,
        **ExtraLayerAttribute.to_kwargs(layer_attr))
Z
zhangjinchao01 已提交
5495

Q
qijun 已提交
5496 5497 5498 5499 5500
    return LayerOutput(
        name,
        LayerType.CROSS_ENTROPY_WITH_SELFNORM,
        parents=[input, label],
        size=1)
Z
zhangjinchao01 已提交
5501

5502

X
xuwei06 已提交
5503 5504 5505 5506 5507 5508
@wrap_name_default()
@layer_support()
def sum_cost(input, name=None, layer_attr=None):
    """
    A loss layer which calculate the sum of the input as loss

C
caoying03 已提交
5509 5510
    The example usage is:

X
xuwei06 已提交
5511 5512
    .. code-block:: python

L
Luo Tao 已提交
5513
       cost = sum_cost(input=input_layer)
X
xuwei06 已提交
5514 5515 5516 5517 5518 5519 5520 5521 5522 5523

    :param input: The first input layer.
    :type input: LayerOutput.
    :param name: The name of this layers. It is not necessary.
    :type name: None|basestring.
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
    :return: LayerOutput object.
    :rtype: LayerOutput.
    """
L
Luo Tao 已提交
5524
    assert isinstance(input, LayerOutput)
Q
qijun 已提交
5525 5526 5527 5528 5529
    Layer(
        name=name,
        type=LayerType.SUM_COST,
        inputs=[input.name],
        **ExtraLayerAttribute.to_kwargs(layer_attr))
X
xuwei06 已提交
5530

Q
qijun 已提交
5531
    return LayerOutput(name, LayerType.SUM_COST, parents=[input], size=1)
X
xuwei06 已提交
5532 5533


Z
zhangjinchao01 已提交
5534
@wrap_name_default()
L
luotao1 已提交
5535 5536
@layer_support()
def huber_cost(input, label, name=None, coeff=1.0, layer_attr=None):
Z
zhangjinchao01 已提交
5537 5538 5539
    """
    A loss layer for huber loss.

C
caoying03 已提交
5540 5541
    The example usage is:

Z
zhangjinchao01 已提交
5542 5543
    .. code-block:: python

X
xuwei06 已提交
5544
       cost = huber_cost(input=input_layer,
L
Luo Tao 已提交
5545
                         label=label_layer)
Z
zhangjinchao01 已提交
5546 5547 5548 5549 5550 5551 5552 5553 5554

    :param input: The first input layer.
    :type input: LayerOutput.
    :param label: The input label.
    :type input: LayerOutput.
    :param name: The name of this layers. It is not necessary.
    :type name: None|basestring.
    :param coeff: The coefficient affects the gradient in the backward.
    :type coeff: float.
L
luotao1 已提交
5555 5556
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
D
dangqingqing 已提交
5557
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
5558 5559
    :rtype: LayerOutput.
    """
5560 5561 5562
    assert isinstance(input, LayerOutput)
    if input.size is not None:
        assert input.size == 1
Q
qijun 已提交
5563 5564 5565 5566 5567 5568
    Layer(
        name=name,
        type=LayerType.HUBER,
        inputs=[input.name, label.name],
        coeff=coeff,
        **ExtraLayerAttribute.to_kwargs(layer_attr))
X
xuwei06 已提交
5569
    return LayerOutput(name, LayerType.HUBER, parents=[input, label], size=1)
Z
zhangjinchao01 已提交
5570

5571

Z
zhangjinchao01 已提交
5572
@wrap_name_default()
L
luotao1 已提交
5573
@layer_support()
Q
qijun 已提交
5574 5575 5576 5577
def multi_binary_label_cross_entropy(input,
                                     label,
                                     name=None,
                                     coeff=1.0,
L
luotao1 已提交
5578
                                     layer_attr=None):
Z
zhangjinchao01 已提交
5579 5580 5581
    """
    A loss layer for multi binary label cross entropy.

C
caoying03 已提交
5582 5583
    The example usage is:

Z
zhangjinchao01 已提交
5584 5585
    .. code-block:: python

X
xuwei06 已提交
5586
       cost = multi_binary_label_cross_entropy(input=input_layer,
L
Luo Tao 已提交
5587
                                               label=label_layer)
Z
zhangjinchao01 已提交
5588 5589 5590 5591 5592 5593 5594 5595 5596

    :param input: The first input layer.
    :type input: LayerOutput
    :param label: The input label.
    :type input: LayerOutput
    :param name: The name of this layers. It is not necessary.
    :type name: None|basestring
    :param coeff: The coefficient affects the gradient in the backward.
    :type coeff: float
L
luotao1 已提交
5597 5598
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
D
dangqingqing 已提交
5599
    :return: LayerOutput object.
Z
zhangjinchao01 已提交
5600 5601 5602
    :rtype: LayerOutput
    """

5603 5604
    if input.activation is None or \
            not isinstance(input.activation, SigmoidActivation):
Q
qijun 已提交
5605 5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616 5617 5618 5619 5620
        logger.log(
            logging.WARN,
            "%s is not recommend for multi_binary_label_cross_entropy's activation, "
            "maybe the sigmoid is better" % repr(input.activation))

    Layer(
        name=name,
        type=LayerType.MULTI_BIN_LABEL_CROSS_ENTROPY,
        inputs=[input.name, label.name],
        coeff=coeff,
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name,
        LayerType.MULTI_BIN_LABEL_CROSS_ENTROPY,
        parents=[input, label],
        size=1)
D
dangqingqing 已提交
5621 5622 5623 5624


@wrap_name_default()
@layer_support()
5625
def smooth_l1_cost(input, label, name=None, coeff=1.0, layer_attr=None):
D
dangqingqing 已提交
5626 5627
    """
    This is a L1 loss but more smooth. It requires that the
D
dangqingqing 已提交
5628
    size of input and label are equal. The formula is as follows,
D
dangqingqing 已提交
5629 5630 5631 5632 5633 5634 5635 5636 5637

    .. math::

        L = \sum_{i} smooth_{L1}(input_i - label_i)

    in which

    .. math::

5638
        smooth_{L1}(x) = \\begin{cases} 0.5x^2& \\text{if}  \\ |x| < 1 \\\\ |x|-0.5& \\text{otherwise} \end{cases}
D
dangqingqing 已提交
5639

D
dangqingqing 已提交
5640 5641 5642
    More details can be found by referring to `Fast R-CNN
    <https://arxiv.org/pdf/1504.08083v2.pdf>`_

C
caoying03 已提交
5643 5644
    The example usage is:

D
dangqingqing 已提交
5645 5646
    .. code-block:: python

5647 5648
       cost = smooth_l1_cost(input=input_layer,
                             label=label_layer)
D
dangqingqing 已提交
5649 5650 5651 5652 5653 5654 5655

    :param input: The input layer.
    :type input: LayerOutput
    :param label: The input label.
    :type input: LayerOutput
    :param name: The name of this layers. It is not necessary.
    :type name: None|basestring
5656 5657
    :param coeff: The coefficient affects the gradient in the backward.
    :type coeff: float
D
dangqingqing 已提交
5658 5659 5660 5661 5662 5663 5664 5665 5666 5667 5668 5669 5670
    :param layer_attr: Extra Layer Attribute.
    :type layer_attr: ExtraLayerAttribute
    :return: LayerOutput object.
    :rtype: LayerOutput
    """
    assert isinstance(input, LayerOutput)
    assert isinstance(label, LayerOutput)
    assert input.size == label.size

    Layer(
        name=name,
        type=LayerType.SMOOTH_L1,
        inputs=[input.name, label.name],
5671
        coeff=coeff,
D
dangqingqing 已提交
5672 5673 5674
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name, LayerType.SMOOTH_L1, parents=[input, label], size=1)
W
wwhu 已提交
5675 5676 5677 5678 5679 5680 5681 5682 5683 5684 5685 5686 5687 5688 5689 5690 5691 5692 5693


@wrap_name_default()
def multiplex_layer(input, name=None, layer_attr=None):
    """
    This layer multiplex multiple layers according to the index,
    which is provided by the first input layer.
    inputs[0]: the index of the layer to output of size batchSize.
    inputs[1:N]; the candidate output data.
    For each index i from 0 to batchSize -1, the output is the i-th row of the
    (index[i] + 1)-th layer.

    For each i-th row of output:
    .. math::
        y[i][j] = x_{x_{0}[i] + 1}[i][j], j = 0,1, ... , (x_{1}.width - 1)

    where, y is output. :math:`x_{k}` is the k-th input layer and
    :math:`k = x_{0}[i] + 1`.

C
caoying03 已提交
5694 5695
    The example usage is:

W
wwhu 已提交
5696 5697 5698 5699 5700 5701 5702 5703 5704 5705 5706 5707 5708 5709 5710 5711 5712 5713 5714 5715 5716 5717 5718 5719 5720 5721 5722 5723 5724 5725 5726 5727
    .. code-block:: python

       maxid = multiplex_layer(input=layers)

    :param input: Input layers.
    :type input: list of LayerOutput
    :param name: Layer name.
    :type name: basestring
    :param layer_attr: extra layer attributes.
    :type layer_attr: ExtraLayerAttribute.
    :return: LayerOutput object.
    :rtype: LayerOutput
    """

    assert isinstance(input, collections.Sequence)
    assert len(input) > 2, 'multiplex_layer should have more than 2 inputs'
    for i in range(1, len(input)):
        assert isinstance(input[i], LayerOutput)
        assert input[i].size == input[1].size, \
            "All the input layers except the first one should have the same size"

    l = Layer(
        name=name,
        type='multiplex',
        inputs=[x.name for x in input],
        size=input[1].size,
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name=name,
        layer_type=LayerType.MULTIPLEX_LAYER,
        parents=input,
        size=l.config.size)
D
dangqingqing 已提交
5728 5729


5730 5731 5732 5733 5734 5735 5736 5737 5738 5739 5740 5741 5742 5743 5744 5745
@wrap_name_default("dropout")
def dropout_layer(input, dropout_rate, name=None):
    """
    @TODO(yuyang18): Add comments.

    :param name:
    :param input:
    :param dropout_rate:
    :return:
    """
    return addto_layer(
        name=name,
        input=input,
        act=LinearActivation(),
        bias_attr=False,
        layer_attr=ExtraAttr(drop_rate=dropout_rate))
5746 5747


D
dangqingqing 已提交
5748 5749 5750 5751 5752 5753 5754 5755 5756 5757 5758 5759 5760 5761 5762 5763 5764 5765 5766 5767 5768 5769
@wrap_name_default()
@wrap_act_default(act=LinearActivation())
@wrap_param_attr_default()
@layer_support(DROPOUT)
def row_conv_layer(input,
                   context_len,
                   act=None,
                   name=None,
                   param_attr=None,
                   layer_attr=None):
    """

    The row convolution is called lookahead convolution. It is firstly
    introduced in paper of `Deep Speech 2: End-toEnd Speech Recognition
    in English and Mandarin <https://arxiv.org/pdf/1512.02595v1.pdf>`_ .

    The bidirectional RNN that learns representation for a sequence by
    performing a forward and a backward pass through the entire sequence.
    However, unlike unidirectional RNNs, bidirectional RNNs are challenging
    to deploy in an online and low-latency setting. The lookahead convolution
    incorporates information from future subsequences in a computationally
    efficient manner to improve unidirectional recurrent neural networks.
5770

D
dangqingqing 已提交
5771 5772 5773 5774 5775
    The connection of row convolution is different form the 1D sequence
    convolution. Assumed that, the future context-length is k, that is to say,
    it can get the output at timestep t by using the the input feature from t-th
    timestep to (t+k+1)-th timestep. Assumed that the hidden dim of input
    activations are d, the activations r_t for the new layer at time-step t are:
5776

D
dangqingqing 已提交
5777 5778 5779 5780 5781 5782 5783 5784 5785 5786 5787 5788 5789 5790 5791 5792 5793 5794 5795 5796 5797 5798 5799 5800 5801 5802 5803 5804 5805 5806 5807 5808 5809 5810 5811 5812 5813 5814 5815 5816 5817 5818 5819
    .. math::

        r_{t,r} = \sum_{j=1}^{k + 1} {w_{i,j}h_{t+j-1, i}}
                  \quad \text{for} \quad  (1 \leq i \leq d)

    Note:
        The `context_len` is `k + 1`. That is to say, the lookahead step
        number plus one equals context_len.


    .. code-block:: python

       row_conv = row_conv_layer(input=input_layer, context_len=3)


    :param input: The input layer.
    :type input: LayerOutput
    :param context_len: The context length equals the lookahead step number
                        plus one.
    :type context_len: int
    :param act: Activation Type. Default is linear activation.
    :type act: BaseActivation
    :param param_attr: The Parameter Attribute. If None, the parameter will be
                       initialized smartly. It's better set it by yourself.
    :type param_attr: ParameterAttribute
    :param layer_attr: Extra Layer config.
    :type layer_attr: ExtraLayerAttribute|None
    :return: LayerOutput object.
    :rtype: LayerOutput

    """
    assert isinstance(input, LayerOutput)
    assert context_len > 0, "the context_len must be greatet than 0."

    Layer(
        inputs=[Input(input.name, **param_attr.attr)],
        name=name,
        context_length=context_len,
        type=LayerType.ROW_CONV_LAYER,
        active_type=act.name,
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name, LayerType.ROW_CONV_LAYER, input, activation=act, size=input.size)
D
dangqingqing 已提交
5820 5821


5822 5823 5824 5825 5826 5827 5828 5829 5830 5831 5832 5833 5834 5835 5836 5837 5838 5839 5840
@layer_support()
@wrap_name_default()
@wrap_param_attr_default()
def prelu_layer(input,
                name=None,
                partial_sum=1,
                param_attr=None,
                layer_attr=None):
    """
    The Parameter Relu activation that actives outputs with a learnable weight.

    Reference:
        Delving Deep into Rectifiers: Surpassing Human-Level Performance on
        ImageNet Classification http://arxiv.org/pdf/1502.01852v1.pdf

    .. math::
       z_i &\\quad if \\quad z_i > 0 \\\\
       a_i * z_i  &\\quad \\mathrm{otherwise}

C
caoying03 已提交
5841 5842 5843 5844 5845 5846
    The example usage is:

    .. code-block:: python

       prelu = prelu_layer(input=layers, partial_sum=1)

5847 5848 5849 5850 5851
    :param name: Name of this layer.
    :type name: basestring
    :param input: The input layer.
    :type input: LayerOutput
    :param partial_sum: this parameter makes a group of inputs share a same weight.
C
caoying03 已提交
5852 5853 5854 5855 5856 5857

        - partial_sum = 1, indicates the element-wise activation: each element has a weight.
        - partial_sum = number of elements in one channel, indicates the channel-wise activation, elements in a channel share a same weight.
        - partial_sum = number of outputs, indicates all elements share a same weight.

    :type partial_sum: int
5858 5859 5860 5861 5862 5863 5864 5865
    :param param_attr: The parameter attribute. See ParameterAttribute for details.
    :type param_attr: ParameterAttribute|None
    :param layer_attr: Extra layer configurations. Default is None.
    :type layer_attr: ExtraLayerAttribute|None
    :return: LayerOutput object.
    :rtype: LayerOutput
    """

5866
    assert isinstance(input, LayerOutput), 'prelu_layer accepts only one input.'
C
caoying03 已提交
5867
    assert isinstance(param_attr, ParameterAttribute)
5868 5869 5870

    l = Layer(
        name=name,
C
caoying03 已提交
5871
        type=LayerType.PRELU,
C
caoying03 已提交
5872
        inputs=Input(input.name, **param_attr.attr),
5873 5874 5875 5876 5877 5878 5879
        partial_sum=partial_sum,
        **ExtraLayerAttribute.to_kwargs(layer_attr))
    return LayerOutput(
        name=name,
        layer_type=LayerType.PRELU,
        parents=input,
        size=l.config.size)
5880 5881 5882


@wrap_name_default()
C
caoying03 已提交
5883
@layer_support(ERROR_CLIPPING, DROPOUT)
5884 5885 5886 5887 5888 5889 5890
@wrap_act_default(act=LinearActivation())
def gated_unit_layer(input,
                     size,
                     act=None,
                     name=None,
                     gate_attr=None,
                     gate_param_attr=None,
C
caoying03 已提交
5891 5892
                     gate_bias_attr=True,
                     inproj_attr=None,
5893 5894 5895 5896 5897 5898 5899 5900 5901 5902 5903 5904 5905 5906 5907 5908 5909 5910 5911 5912 5913 5914 5915 5916 5917 5918 5919 5920 5921 5922 5923 5924 5925 5926 5927 5928
                     inproj_param_attr=None,
                     inproj_bias_attr=True,
                     layer_attr=None):
    """
    The gated unit layer implements a simple gating mechanism over the input.
    The input :math:`X` is first projected into a new space :math:`X'`, and
    it is also used to produce a gate weight :math:`\sigma`. Element-wise
    prodict between :match:`X'` and :math:`\sigma` is finally returned.

    Reference:
        Language Modeling with Gated Convolutional Networks
        https://arxiv.org/abs/1612.08083

    .. math::
       y=\\text{act}(X \cdot W + b)\otimes \sigma(X \cdot V + c)

    The example usage is:

    .. code-block:: python
        gated_unit = gated_unit_layer(size=128, input=input_layer))

    :param input: input for this layer.
    :type input: LayerOutput
    :param size: output size of the gated unit.
    :type size: int
    :param act: activation type of the projected input.
    :type act: BaseActivation
    :param name: name of this layer.
    :type name: basestring
    :param gate_attr: Attributes to tune the gate output, for example, error
        clipping threshold, dropout and so on. See ExtraLayerAttribute for
        more details.
    :type gate_attr: ExtraLayerAttribute|None
    :param gate_param_attr: Attributes to tune the learnable projected matrix
        parameter of the gate.
    :type gate_param_attr: ParameterAttribute|None
C
caoying03 已提交
5929 5930 5931 5932 5933 5934
    :param gate_bias_attr: Attributes to tune the learnable bias of the gate.
    :type gate_bias_attr: ParameterAttribute|None
    :param inproj_attr: Attributes to the tune the projected input, for
        example, error clipping threshold, dropout and so on. See
        ExtraLayerAttribute for more details.
    :type inproj_attr: ExtraLayerAttribute|None
5935 5936 5937 5938 5939 5940 5941 5942 5943 5944 5945 5946 5947 5948 5949 5950 5951 5952 5953 5954 5955 5956
    :param inproj_param_attr: Attributes to tune the learnable parameter of
        the projection of input.
    :type inproj_param_attr: ParameterAttribute|None
    :param inproj_bias_attr: Attributes to tune the learnable bias of
        projection of the input.
    :type inproj_bias_attr: ParameterAttribute|None
    :param layer_attr: Attributes to tune the final output of the gated unit,
        for example, error clipping threshold, dropout and so on. See
        ExtraLayerAttribute for more details.
    :type layer_attr: ExtraLayerAttribute|None
    :return: LayerOutput object.
    :rtype: LayerOutput
    """

    assert isinstance(
        input, LayerOutput), 'The gated linear unit accepts only one input.'

    input_proj = fc_layer(
        input=input,
        name="%s_input_proj" % name,
        size=size,
        act=act,
C
caoying03 已提交
5957
        layer_attr=inproj_attr,
5958 5959 5960 5961 5962 5963 5964 5965 5966
        param_attr=inproj_param_attr,
        bias_attr=inproj_bias_attr)

    gate = fc_layer(
        size=size,
        name="%s_gate" % name,
        act=SigmoidActivation(),
        input=input,
        layer_attr=gate_attr,
C
caoying03 已提交
5967
        param_attr=gate_param_attr,
5968 5969 5970 5971 5972
        bias_attr=gate_bias_attr)
    return mixed_layer(
        name="%s_gated_act" % name,
        input=dotmul_operator(input_proj, gate),
        layer_attr=layer_attr)