tensor.py 15.3 KB
Newer Older
1
#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
D
dzhwinter 已提交
2
#
D
dzhwinter 已提交
3 4 5
# 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
D
dzhwinter 已提交
6
#
D
dzhwinter 已提交
7
#     http://www.apache.org/licenses/LICENSE-2.0
D
dzhwinter 已提交
8
#
D
dzhwinter 已提交
9 10 11 12 13 14
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

Y
Yu Yang 已提交
15
from ..layer_helper import LayerHelper
16
from ..param_attr import ParamAttr
X
xuwei06 已提交
17 18
from ..framework import convert_np_dtype_to_dtype_
from ..framework import Variable
19
from ..initializer import Constant, force_init_on_cpu
20
from ..core import VarDesc
Y
yuyang18 已提交
21
from layer_function_generator import templatedoc
X
xuwei06 已提交
22
import numpy
Y
Yu Yang 已提交
23 24

__all__ = [
25 26
    'create_tensor',
    'create_parameter',
Q
Qiao Longfei 已提交
27
    'create_global_var',
28 29 30 31 32 33
    'cast',
    'concat',
    'sums',
    'assign',
    'fill_constant_batch_size_like',
    'fill_constant',
S
sneaxiy 已提交
34 35
    'argmin',
    'argmax',
36 37
    'ones',
    'zeros',
Y
Yu Yang 已提交
38 39 40
]


X
xuwei06 已提交
41
def create_tensor(dtype, name=None, persistable=False):
Y
Yu Yang 已提交
42
    helper = LayerHelper("create_tensor", **locals())
X
xuwei06 已提交
43 44
    return helper.create_variable(
        name=helper.name, dtype=dtype, persistable=persistable)
Y
Yu Yang 已提交
45 46


47 48
def create_parameter(shape,
                     dtype,
X
xuwei06 已提交
49
                     name=None,
50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67
                     attr=None,
                     is_bias=False,
                     default_initializer=None):
    """
    Create a parameter
    Args:
        shape(list[int]): shape of the parameter
        dtype(string): element type of the parameter
        attr(ParamAttr): attributes of the parameter
        is_bias(bool): This can affect which default initializer is chosen
                       when default_initializer is None. If is_bias,
                       initializer.Constant(0.0) will be used. Otherwise,
                       Xavier() will be used.
        default_initializer(Initializer): initializer for the parameter

    Returns:
        Parameter: the created parameter
    """
Q
Qiao Longfei 已提交
68
    helper = LayerHelper("create_parameter", **locals())
69
    if attr is None:
X
xuwei06 已提交
70
        attr = ParamAttr(name=name)
71 72 73 74
    return helper.create_parameter(attr, shape, dtype, is_bias,
                                   default_initializer)


75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92
def create_global_var(shape,
                      value,
                      dtype,
                      persistable=False,
                      force_cpu=False,
                      name=None):
    """
    Create a global variable. such as global_step
    Args:
        shape(list[int]): shape of the variable
        value(float): the value of the variable
        dtype(string): element type of the parameter
        persistable(bool): if this variable is persistable
        force_cpu(bool): force this variable to be on CPU

    Returns:
        Variable: the created Variable
    """
Q
Qiao Longfei 已提交
93 94 95 96
    helper = LayerHelper("global_var", **locals())
    var = helper.create_global_variable(
        dtype=dtype, shape=shape, persistable=persistable, name=name)
    helper.set_variable_initializer(
97 98
        var, initializer=Constant(
            value=float(value), force_cpu=force_cpu))
Q
Qiao Longfei 已提交
99 100 101
    return var


102
def cast(x, dtype):
Y
Yu Yang 已提交
103 104 105 106 107 108 109 110 111 112 113 114 115 116 117
    """
    This function takes in the input with input_dtype
    and casts it to the output_dtype as the output.
    """
    helper = LayerHelper('cast', **locals())
    out = helper.create_tmp_variable(dtype=dtype)
    helper.append_op(
        type='cast',
        inputs={'X': [x]},
        outputs={'Out': [out]},
        attrs={'in_dtype': x.dtype,
               'out_dtype': out.dtype})
    return out


118
def concat(input, axis=0, name=None):
Y
Yu Yang 已提交
119
    """
120 121 122
    **Concat**

    This function concatenates the input along the axis mentioned
Y
Yu Yang 已提交
123
    and returns that as the output.
124 125 126 127

    Args:
        input(list): List of tensors to be concatenated
        axis(int): Integer axis along which the tensors will be concatenated
128 129
        name(str|None): A name for this layer(optional). If set None, the layer
                       will be named automatically.
130 131 132 133 134 135 136

    Returns:
        Variable: Output variable of the concatenation

    Examples:
        .. code-block:: python
          out = fluid.layers.concat(input=[Efirst, Esecond, Ethird, Efourth])
Y
Yu Yang 已提交
137 138 139 140 141 142 143 144 145 146 147
    """
    helper = LayerHelper('concat', **locals())
    out = helper.create_tmp_variable(dtype=helper.input_dtype())
    helper.append_op(
        type='concat',
        inputs={'X': input},
        outputs={'Out': [out]},
        attrs={'axis': axis})
    return out


148
def sums(input, out=None):
K
kavyasrinet 已提交
149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167
    """This function performs the sum operation on the input and returns the
    result as the output.

    Args:
        input (Variable|list): The input tensor that has the elements
                               that need to be summed up.

    Returns:
        Variable: The tensor type variable that has the sum of input
                  written to it.

    Examples:
        .. code-block::python

          tmp = fluid.layers.zeros(shape=[10], dtype='int32')
          i = fluid.layers.fill_constant(shape=[1], dtype='int64', value=10)
          a0 = layers.array_read(array=tmp, i=i)
          i = layers.increment(x=i)
          a1 = layers.array_read(array=tmp, i=i)
Y
Yu Yang 已提交
168 169
          mean_a0 = layers.mean(a0)
          mean_a1 = layers.mean(a1)
K
kavyasrinet 已提交
170
          a_sum = layers.sums(input=[mean_a0, mean_a1])
Y
Yu Yang 已提交
171 172 173 174 175 176 177 178
    """
    helper = LayerHelper('sum', **locals())
    if out is None:
        out = helper.create_tmp_variable(dtype=helper.input_dtype())
    helper.append_op(type='sum', inputs={'X': input}, outputs={'Out': out})
    return out


179
def assign(input, output):
180 181 182 183 184 185
    """
    **Assign**

    This function copies the *input* Variable to the *output* Variable.

    Args:
X
xuwei06 已提交
186
        input(Variable|numpy.ndarray): The source variable
187 188 189 190 191 192 193 194 195 196 197
        output(Variable): The destination variable

    Returns:
        Variable: The destination variable that was supplied as the *output*.

    Examples:
        .. code-block:: python
          out = fluid.layers.create_tensor(dtype='float32')
          hidden = fluid.layers.fc(input=data, size=10)
          fluid.layers.assign(hidden, out)
    """
Y
Yu Yang 已提交
198
    helper = LayerHelper('assign', **locals())
X
xuwei06 已提交
199 200
    if isinstance(input, Variable):
        helper.append_op(
R
robot 已提交
201
            type='assign', inputs={'X': [input]}, outputs={'Out': [output]})
X
xuwei06 已提交
202 203
    elif isinstance(input, numpy.ndarray):
        dtype = convert_np_dtype_to_dtype_(input.dtype)
204
        if dtype == VarDesc.VarType.FP32:
X
xuwei06 已提交
205
            value_name = "fp32_values"
206
            values = [float(v) for v in input.flat]
207
        elif dtype == VarDesc.VarType.INT32:
X
xuwei06 已提交
208
            value_name = "int32_values"
209
            values = [int(v) for v in input.flat]
X
xuwei06 已提交
210 211
        else:
            raise ValueError("Unsupported dtype %s", input.dtype)
212 213 214
        if input.size > 1024 * 1024:
            raise ValueError("The size of input is too big. Please consider "
                             "saving it to file and 'load_op' to load it")
X
xuwei06 已提交
215 216 217 218 219 220 221

        helper.append_op(
            type='assign_value',
            outputs={'Out': [output]},
            attrs={
                'dtype': dtype,
                'shape': list(input.shape),
222
                value_name: values
X
xuwei06 已提交
223 224 225 226
            })
    else:
        raise ValueError("Wrong type for assign input: %s" % type(input))

Y
Yu Yang 已提交
227 228 229
    return output


Q
QI JUN 已提交
230
def fill_constant(shape, dtype, value, force_cpu=False, out=None):
Y
Yu Yang 已提交
231
    """
232 233
    **fill_constant**

234 235
    This function creates a tensor with specified `shape` and `dtype`, and
    initializes it with a constant specifed by `value`.
K
kavyasrinet 已提交
236

237
    The attribute `stop_gradient` of the created tensor is set to True.
238 239

    Args:
240
        shape(tuple|list|None): Shape of the output tensor.
241
        dtype(np.dtype|core.VarDesc.VarType|str): Data type of the output tensor.
242 243
        value(float): The constant value used to initialize the output tensor.
        out(Variable): The output tensor.
244
        force_cpu(True|False): data should be on CPU if set true.
245 246

    Returns:
247
        Variable: The tensor variable storing the output.
248 249 250 251 252

    Examples:
        .. code-block:: python

          data = fluid.layers.fill_constant(shape=[1], value=0, dtype='int64')
Y
Yu Yang 已提交
253
    """
254

Y
Yu Yang 已提交
255 256 257 258 259 260 261
    helper = LayerHelper("fill_constant", **locals())
    if out is None:
        out = helper.create_tmp_variable(dtype=dtype)
    helper.append_op(
        type='fill_constant',
        inputs={},
        outputs={'Out': [out]},
Q
QI JUN 已提交
262 263 264 265
        attrs={
            'shape': shape,
            'dtype': out.dtype,
            'value': float(value),
266
            'force_cpu': force_cpu or force_init_on_cpu()
Q
QI JUN 已提交
267
        })
Y
Yu Yang 已提交
268 269 270 271
    out.stop_gradient = True
    return out


Y
yuyang18 已提交
272
@templatedoc()
Y
Yu Yang 已提交
273 274 275 276 277
def fill_constant_batch_size_like(input,
                                  shape,
                                  dtype,
                                  value,
                                  input_dim_idx=0,
278
                                  output_dim_idx=0):
279
    """
Y
yuyang18 已提交
280
    ${comment}
281 282 283

    It also sets *stop_gradient* to True.

Y
yuyang18 已提交
284 285 286
    >>> data = fluid.layers.fill_constant_batch_size_like(
    >>>             input=like, shape=[1], value=0, dtype='int64')

287
    Args:
Y
yuyang18 已提交
288
        input(${input_type}): ${input_comment}.
289

Y
yuyang18 已提交
290
        shape(${shape_type}): ${shape_comment}.
291

Y
yuyang18 已提交
292 293 294
        dtype(${dtype_type}): ${dtype_comment}.

        value(${value_type}): ${value_comment}.
295

Y
yuyang18 已提交
296 297 298 299 300
        input_dim_idx(${input_dim_idx_type}): ${input_dim_idx_comment}.

        output_dim_idx(${output_dim_idx_type}): ${output_dim_idx_comment}.

    Returns:
Y
yuyang18 已提交
301
        ${out_comment}.
302
    """
Y
Yu Yang 已提交
303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319
    helper = LayerHelper("fill_constant_batch_size_like", **locals())
    out = helper.create_tmp_variable(dtype=dtype)
    helper.append_op(
        type='fill_constant_batch_size_like',
        inputs={'Input': input},
        outputs={'Out': [out]},
        attrs={
            'shape': shape,
            'dtype': out.dtype,
            'value': float(value),
            'input_dim_idx': input_dim_idx,
            'output_dim_idx': output_dim_idx
        })
    out.stop_gradient = True
    return out


S
sneaxiy 已提交
320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381
def argmin(x, axis=0):
    """
    **argmin**

    This function computes the indices of the min elements 
    of the input tensor's element along the provided axis.

    Args:
        x(Variable): The input to compute the indices of
                     the min elements.
        axis(int): Axis to compute indices along.
    
    Returns:
        Variable: The tensor variable storing the output
    
    Examples:
        .. code-block:: python
          
          out = fluid.layers.argmin(x=in, axis=0)
          out = fluid.layers.argmin(x=in, axis=-1)  
    """
    helper = LayerHelper("arg_min", **locals())
    out = helper.create_tmp_variable(VarDesc.VarType.INT64)
    helper.append_op(
        type='arg_min',
        inputs={'X': x},
        outputs={'Out': [out]},
        attrs={'axis': axis})
    return out


def argmax(x, axis=0):
    """
    **argmax**

    This function computes the indices of the max elements 
    of the input tensor's element along the provided axis.

    Args:
        x(Variable): The input to compute the indices of
                     the max elements.
        axis(int): Axis to compute indices along.
    
    Returns:
        Variable: The tensor variable storing the output
    
    Examples:
        .. code-block:: python
          
          out = fluid.layers.argmax(x=in, axis=0)
          out = fluid.layers.argmax(x=in, axis=-1)  
    """
    helper = LayerHelper("arg_max", **locals())
    out = helper.create_tmp_variable(VarDesc.VarType.INT64)
    helper.append_op(
        type='arg_max',
        inputs={'X': x},
        outputs={'Out': [out]},
        attrs={'axis': axis})
    return out


Y
Yang Yu 已提交
382
def ones(shape, dtype, force_cpu=False):
Y
Yu Yang 已提交
383
    """
384 385 386 387 388 389 390 391 392
    **ones**

    This function creates a tensor of specified *shape* and
    *dtype*, and initializes this with 1.

    It also sets *stop_gradient* to True.

    Args:
        shape(tuple|list|None): Shape of output tensor
393
        dtype(np.dtype|core.VarDesc.VarType|str): Data type of output tensor
394 395 396 397 398 399 400 401

    Returns:
        Variable: The tensor variable storing the output

    Examples:
        .. code-block:: python

          data = fluid.layers.ones(shape=[1], dtype='int64')
Y
Yu Yang 已提交
402 403 404 405
    """
    return fill_constant(value=1.0, **locals())


Y
Yang Yu 已提交
406
def zeros(shape, dtype, force_cpu=False):
Y
Yu Yang 已提交
407
    """
408 409 410 411 412 413 414 415 416
    **zeros**

    This function creates a tensor of specified *shape* and
    *dtype*, and initializes this with 0.

    It also sets *stop_gradient* to True.

    Args:
        shape(tuple|list|None): Shape of output tensor
417
        dtype(np.dtype|core.VarDesc.VarType|str): Data type of output tensor
418 419 420 421 422 423 424 425

    Returns:
        Variable: The tensor variable storing the output

    Examples:
        .. code-block:: python

          data = fluid.layers.zeros(shape=[1], dtype='int64')
Y
Yu Yang 已提交
426 427
    """
    return fill_constant(value=0.0, **locals())
428 429


F
fengjiayi 已提交
430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463
def reverse(x, axis):
    """
    **reverse**

    This function reverse the input 'x' along given axises.

    Args:
        x(Vairbale): the input to be reversed.
        axis(int|tuple|list): Axis that along which order of elements 
                    is reversed. If it is a tuple or a list, reversing 
                    will be apply on each axis in the tuple or list.  

    Returns:
        Variable: The reversed tensor.

    Examples:
        .. code-block:: python

          out = fluid.layers.reverse(x=in, axis=0)
          # or:
          out = fluid.layers.reverse(x=in, axis=[0,1])
    """
    if isinstance(axis, int):
        axis = [axis]
    helper = LayerHelper("reverse", **locals())
    out = helper.create_tmp_variable(dtype=x.dtype)
    helper.append_op(
        type='reverse',
        inputs={'Input': x},
        outputs={'Out': [out]},
        attrs={'axis': axis})
    return out


464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517
def save(x, file_path, overwrite=True):
    """
    Saves a variable as a file.

    Args:
        x(variable): The Tensor/LoDTensor to be saved.
        file_path(str): The file path where the variable will be saved.
        overwrite(bool): Whether or not cover the given file when it has already 
            existed. If it's set 'False' and the file is existed, a runtime 
            error will be thrown. 
    """
    helper = LayerHelper("save", **locals())
    helper.append_op(
        type="save",
        inputs={"input": x},
        outputs={},
        args={"file_path": file_path,
              "overwrite": overwrite})


def save_combine(x, file_path, overwrite=True):
    """
    Saves a list of variables into a single file.

    Args:
        x(list): A list of Tensor/LoDTensor to be saved together in a single file.
        file_path(str): The file path where variables will be saved.
        overwrite(bool): Whether or not cover the given file when it has already 
            existed. If it's set 'False' and the file is existed, a runtime 
            error will be thrown. 
    """
    helper = LayerHelper("save_combine", **locals())
    helper.append_op(
        type="save_combine",
        inputs={"input": x},
        outputs={},
        args={"file_path": file_path,
              "overwrite": overwrite})


def load_combine(out, file_path):
    """
    Loads a list of vairables from a single file.

    Args:
        out(list): The list of variables to be read from the disk file.
        file_path(str): The path of the disk file.
    """
    helper = LayerHelper("load_combine", **locals())
    helper.append_op(
        type="load_combine",
        inputs={},
        output={"Out": out},
        args={"file_path": file_path})