layers.py 29.2 KB
Newer Older
Y
Yu Yang 已提交
1
import paddle.v2.framework.core as core
Y
Yu Yang 已提交
2 3 4 5
from paddle.v2.framework.framework import OpProtoHolder, Variable, Program, \
    Operator
from paddle.v2.framework.initializer import ConstantInitializer, \
    NormalInitializer
6
from paddle.v2.framework.layer_helper import LayerHelper, unique_name
Y
Yu Yang 已提交
7 8
import re

Q
QI JUN 已提交
9
__all__ = [
Y
Yu Yang 已提交
10
    'fc', 'data', 'cross_entropy', 'conv2d', 'pool2d', 'embedding', 'concat',
D
dzhwinter 已提交
11 12
    'StaticRNN', 'cast', 'sequence_conv', 'sequence_pool', 'sums', 'cos_sim',
    'batch_norm', 'accuracy'
Q
QI JUN 已提交
13
]
Y
Yu Yang 已提交
14 15


F
fengjiayi 已提交
16 17 18 19 20 21 22
def fc(input,
       size,
       param_attr=None,
       bias_attr=True,
       name=None,
       act=None,
       num_flatten_dims=1,
23 24
       main_program=None,
       startup_program=None):
Y
Yu Yang 已提交
25 26 27 28 29 30 31 32 33
    # create helper
    helper = LayerHelper('fc', **locals())

    dtype = helper.input_dtype()

    # mul
    mul_results = []
    for input_var, param_attr in helper.iter_inputs_and_params():
        input_shape = input_var.shape
Y
Yu Yang 已提交
34 35 36
        param_shape = [
            reduce(lambda a, b: a * b, input_shape[num_flatten_dims:], 1)
        ] + [size]
Y
Yu Yang 已提交
37 38 39 40 41 42 43 44 45 46
        w = helper.create_parameter(
            attr=param_attr, shape=param_shape, dtype=dtype)
        tmp = helper.create_tmp_variable(dtype)
        helper.append_op(
            type="mul",
            inputs={
                "X": input_var,
                "Y": w,
            },
            outputs={"Out": tmp},
Y
Yu Yang 已提交
47 48
            attrs={'x_num_col_dims': num_flatten_dims,
                   'y_num_col_dims': 1})
Y
Yu Yang 已提交
49 50 51 52 53 54 55 56 57 58 59 60 61 62 63
        mul_results.append(tmp)

    # sum
    if len(mul_results) == 1:
        pre_bias = mul_results[0]
    else:
        pre_bias = helper.create_tmp_variable(dtype)
        helper.append_op(
            type="sum", inputs={"X": mul_results}, outputs={"Out": pre_bias})
    # add bias
    pre_activation = helper.append_bias_op(pre_bias)
    # add activation
    return helper.append_activation(pre_activation)


Q
QI JUN 已提交
64 65 66
def embedding(input,
              size,
              data_type='float32',
67
              is_sparse=False,
Q
QI JUN 已提交
68
              param_attr=None,
69 70
              main_program=None,
              startup_program=None):
Q
QI JUN 已提交
71 72 73 74 75 76 77 78
    helper = LayerHelper('embedding', **locals())
    w = helper.create_parameter(
        attr=helper.param_attr, shape=size, dtype=data_type)
    tmp = helper.create_tmp_variable(data_type)
    helper.append_op(
        type='lookup_table',
        inputs={'Ids': input,
                'W': w},
79 80
        outputs={'Out': tmp},
        attrs={'is_sparse': is_sparse})
Q
QI JUN 已提交
81 82 83
    return tmp


F
fengjiayi 已提交
84 85 86 87
def data(name,
         shape,
         data_type='float32',
         type=core.VarDesc.VarType.LOD_TENSOR,
Y
Yu Yang 已提交
88
         append_batch_size=True,
89 90
         main_program=None,
         startup_program=None):
Y
Yu Yang 已提交
91
    helper = LayerHelper('data', **locals())
Y
Yu Yang 已提交
92 93 94 95 96 97 98 99
    shape = list(shape)
    for i in xrange(len(shape)):
        if shape[i] is None:
            shape[i] = -1
            append_batch_size = False
        elif shape[i] < 0:
            append_batch_size = False

Y
Yu Yang 已提交
100 101
    if append_batch_size:
        shape = [-1] + shape  # append batch size as -1
Y
Yu Yang 已提交
102

Y
Yu Yang 已提交
103
    return helper.create_global_variable(
Y
Yu Yang 已提交
104
        name=name, shape=shape, dtype=data_type, type=type, stop_gradient=True)
Y
Yu Yang 已提交
105 106 107 108 109 110 111 112 113


def _convert_(name):
    s1 = re.sub('(.)([A-Z][a-z]+)', r'\1_\2', name)
    return re.sub('([a-z0-9])([A-Z])', r'\1_\2', s1).lower()


def _create_op_func_(op_type):
    op_proto = OpProtoHolder.instance().get_op_proto(op_type)
114 115 116 117 118 119
    not_intermediate_outputs = \
        filter(lambda output: not output.intermediate, op_proto.outputs)
    intermediate_outputs = \
        filter(lambda output: output.intermediate, op_proto.outputs)

    if len(not_intermediate_outputs) != 1:
Y
Yu Yang 已提交
120
        raise ValueError(
121 122
            "Only one not intermediate output operator can be automatically generated"
        )
Y
Yu Yang 已提交
123

124
    if not_intermediate_outputs[0].duplicable:
Y
Yu Yang 已提交
125 126 127
        raise ValueError(
            "Only not duplicable op can be automatically generated")

128 129 130 131 132 133 134 135
    for output in intermediate_outputs:
        if output.duplicable:
            raise ValueError(
                "Only when all intermediate ops are not duplicable, "
                "this op can be automatically generated")

    o_name = not_intermediate_outputs[0].name
    intermediate_output_names = [output.name for output in intermediate_outputs]
Y
Yu Yang 已提交
136

Y
Yang Yang(Tony) 已提交
137
    def infer_and_check_data_type(op_proto, **kwargs):
Y
Yu Yang 已提交
138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153
        dtype = None
        for ipt in op_proto.inputs:
            name = _convert_(ipt.name)
            val = kwargs.pop(name, [])
            if not isinstance(val, list) and not isinstance(val, tuple):
                val = [val]
            for each in val:
                if not isinstance(each, Variable):
                    raise ValueError("input of {0} must be variable".format(
                        op_type))

                if dtype is None:
                    dtype = each.data_type
                elif dtype != each.data_type:
                    raise ValueError(
                        "operator {0} must input same dtype".format(op_type))
Y
Yang Yang(Tony) 已提交
154 155 156 157 158 159 160 161 162 163 164 165 166 167

        return dtype

    def func(**kwargs):
        helper = LayerHelper(op_type, **kwargs)

        dtype = infer_and_check_data_type(op_proto, **kwargs)

        inputs = dict()
        for ipt in op_proto.inputs:
            name = _convert_(ipt.name)
            val = kwargs.pop(name, [])
            if not isinstance(val, list) and not isinstance(val, tuple):
                val = [val]
Y
Yu Yang 已提交
168 169
            inputs[ipt.name] = val

170
        outputs = dict()
Y
Yu Yang 已提交
171
        out = helper.create_tmp_variable(dtype=dtype)
172 173 174
        outputs[o_name] = [out]
        for name in intermediate_output_names:
            outputs[name] = [helper.create_tmp_variable(dtype=dtype)]
Y
Yu Yang 已提交
175
        helper.append_op(
176
            type=op_type, inputs=inputs, outputs=outputs, attrs=kwargs)
Q
Qiao Longfei 已提交
177
        return helper.append_activation(out)
Y
Yu Yang 已提交
178 179 180 181 182 183 184 185

    func.__name__ = op_type
    globals()[op_type] = func
    global __all__
    __all__.append(op_type)


_create_op_func_('mean')
Y
Yu Yang 已提交
186
_create_op_func_('mul')
Q
Qiao Longfei 已提交
187
_create_op_func_('elementwise_add')
188
_create_op_func_('dropout')
Q
Qiao Longfei 已提交
189
_create_op_func_('reshape')
Y
Yu Yang 已提交
190 191 192
_create_op_func_('elementwise_add')
_create_op_func_('sigmoid')
_create_op_func_('scale')
Y
Yang Yang(Tony) 已提交
193 194 195 196 197 198 199 200 201 202 203 204 205 206
_create_op_func_('reshape')
_create_op_func_('transpose')


def fill_constant(data_type, shape, value=None, program=None):
    helper = LayerHelper('fill_constant', **locals())
    out = helper.create_tmp_variable(dtype=data_type)
    helper.append_op(
        type='fill_constant',
        outputs={'Out': [out]},
        attrs={'data_type': data_type,
               'shape': shape,
               'value': value})
    return out
Y
Yu Yang 已提交
207 208


209
def cast(x, data_type, main_program=None):
Y
Yu Yang 已提交
210 211 212 213 214 215 216 217 218 219 220
    helper = LayerHelper('cast', **locals())
    out = helper.create_tmp_variable(dtype=data_type)
    helper.append_op(
        type='cast',
        inputs={'X': [x]},
        outputs={'Out': [out]},
        attrs={'in_data_type': x.data_type,
               'out_data_type': out.data_type})
    return out


221
def concat(input, axis, main_program=None, startup_program=None):
Q
QI JUN 已提交
222
    helper = LayerHelper('concat', **locals())
D
dzhwinter 已提交
223
    out = helper.create_tmp_variable(dtype=helper.input_dtype())
Q
QI JUN 已提交
224 225 226 227 228 229 230 231
    helper.append_op(
        type='concat',
        inputs={'X': input},
        outputs={'Out': [out]},
        attrs={'axis': axis})
    return out


232
def sums(input, main_program=None, startup_program=None):
D
dzhwinter 已提交
233 234
    helper = LayerHelper('sum', **locals())
    out = helper.create_tmp_variable(dtype=helper.input_dtype())
Y
Yu Yang 已提交
235
    helper.append_op(type='sum', inputs={'X': input}, outputs={'Out': out})
D
dzhwinter 已提交
236 237 238
    return out


239 240 241 242 243
def cos_sim(X, Y, **kwargs):
    helper = LayerHelper('cos_sim', **kwargs)
    out = helper.create_tmp_variable(dtype=X.data_type)
    xnorm = helper.create_tmp_variable(dtype=X.data_type)
    ynorm = helper.create_tmp_variable(dtype=X.data_type)
D
dzhwinter 已提交
244 245 246 247 248 249 250
    helper.append_op(
        type='cos_sim',
        inputs={'X': [X],
                'Y': [Y]},
        outputs={'Out': [out],
                 'XNorm': [xnorm],
                 'YNorm': [ynorm]})
251
    return out
D
dzhwinter 已提交
252 253


Y
Yu Yang 已提交
254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276
def cross_entropy(input, label, **kwargs):
    helper = LayerHelper('cross_entropy', **kwargs)
    out = helper.create_tmp_variable(dtype=input.data_type)
    helper.append_op(
        type='cross_entropy',
        inputs={'X': [input],
                'Label': [label]},
        outputs={'Y': [out]},
        attrs=kwargs)
    return out


def square_error_cost(input, label, **kwargs):
    helper = LayerHelper('square_error_cost', **kwargs)
    minus_out = helper.create_tmp_variable(dtype=input.data_type)
    helper.append_op(
        type='elementwise_sub',
        inputs={'X': [input],
                'Y': [label]},
        outputs={'Out': [minus_out]})

    square_out = helper.create_tmp_variable(dtype=input.data_type)
    helper.append_op(
Q
QI JUN 已提交
277
        type='square', inputs={'X': [minus_out]}, outputs={'Y': [square_out]})
Y
Yu Yang 已提交
278
    return square_out
279 280


F
fengjiayi 已提交
281 282 283 284 285 286 287 288 289 290 291 292 293 294
def accuracy(input, label, k=1, **kwargs):
    helper = LayerHelper("accuracy", **kwargs)
    topk_out = helper.create_tmp_variable(dtype=input.data_type)
    topk_indices = helper.create_tmp_variable(dtype="int64")
    helper.append_op(
        type="top_k",
        inputs={"X": [input]},
        outputs={"Out": [topk_out],
                 "Indices": [topk_indices]},
        attrs={"k": k})
    acc_out_dtype = kwargs.get("out_dtype", "float32")
    acc_out = helper.create_tmp_variable(dtype=acc_out_dtype)
    helper.append_op(
        type="accuracy",
武毅 已提交
295 296 297 298 299
        inputs={
            "Out": [topk_out],
            "Indices": [topk_indices],
            "Label": [label]
        },
F
fengjiayi 已提交
300 301 302 303
        outputs={"Accuracy": [acc_out]})
    return acc_out


D
dzhwinter 已提交
304 305 306
def sequence_conv(input,
                  num_filters,
                  filter_size=3,
307
                  filter_stride=1,
308
                  act=None,
D
dzhwinter 已提交
309 310 311
                  padding=None,
                  bias_attr=None,
                  param_attr=None,
312 313
                  main_program=None,
                  startup_program=None):
D
dzhwinter 已提交
314 315 316 317 318 319 320
    # FIXME(dzh) : want to unify the argument of python layer
    # function. So we ignore some unecessary attributes.
    # such as, padding_trainable, context_start.

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

D
dzhwinter 已提交
321
    filter_shape = [filter_size * input.shape[1], num_filters]
D
dzhwinter 已提交
322 323 324 325 326 327 328 329
    filter = helper.create_parameter(
        attr=helper.param_attr, shape=filter_shape, dtype=dtype)
    pre_bias = helper.create_tmp_variable(dtype)

    helper.append_op(
        type='sequence_conv',
        inputs={
            'X': [input],
D
dzhwinter 已提交
330
            'Filter': [filter],
D
dzhwinter 已提交
331 332 333
        },
        outputs={"Out": pre_bias},
        attrs={
334
            'contextStride': filter_stride,
335
            'contextStart': -int(filter_size / 2),
336
            'contextLength': filter_size
D
dzhwinter 已提交
337 338 339 340 341
        })
    pre_act = helper.append_bias_op(pre_bias)
    return helper.append_activation(pre_act)


F
fengjiayi 已提交
342 343 344 345 346 347 348 349 350 351
def conv2d(input,
           num_filters,
           name=None,
           filter_size=[1, 1],
           act=None,
           groups=None,
           stride=[1, 1],
           padding=None,
           bias_attr=None,
           param_attr=None,
352 353
           main_program=None,
           startup_program=None):
354 355 356 357 358 359 360 361 362 363 364
    helper = LayerHelper('conv2d', **locals())
    dtype = helper.input_dtype()

    num_channels = input.shape[1]
    if groups is None:
        num_filter_channels = num_channels
    else:
        if num_channels % groups is not 0:
            raise ValueError("num_channels must be divisible by groups.")
        num_filter_channels = num_channels / groups

F
fengjiayi 已提交
365 366 367 368 369 370 371
    if isinstance(filter_size, int):
        filter_size = [filter_size, filter_size]
    if isinstance(stride, int):
        stride = [stride, stride]
    if isinstance(padding, int):
        padding = [padding, padding]

372 373
    input_shape = input.shape
    filter_shape = [num_filters, num_filter_channels] + filter_size
374 375

    std = (2.0 / (filter_size[0]**2 * num_channels))**0.5
376
    filter = helper.create_parameter(
377 378 379 380
        attr=helper.param_attr,
        shape=filter_shape,
        dtype=dtype,
        initializer=NormalInitializer(0.0, std, 0))
381 382 383 384 385 386 387 388 389 390 391 392 393
    pre_bias = helper.create_tmp_variable(dtype)

    helper.append_op(
        type='conv2d',
        inputs={
            'Input': input,
            'Filter': filter,
        },
        outputs={"Output": pre_bias},
        attrs={'strides': stride,
               'paddings': padding,
               'groups': groups})

Y
Yu Yang 已提交
394
    pre_act = helper.append_bias_op(pre_bias, 1)
395 396

    return helper.append_activation(pre_act)
F
fengjiayi 已提交
397 398


D
dzhwinter 已提交
399
def sequence_pool(input, pool_type, **kwargs):
400
    helper = LayerHelper('sequence_pool', input=input, **kwargs)
D
dzhwinter 已提交
401 402
    dtype = helper.input_dtype()
    pool_out = helper.create_tmp_variable(dtype)
D
dangqingqing 已提交
403
    max_index = helper.create_tmp_variable(dtype)
D
dzhwinter 已提交
404 405 406

    helper.append_op(
        type="sequence_pool",
D
dangqingqing 已提交
407 408 409
        inputs={"X": input},
        outputs={"Out": pool_out,
                 "MaxIndex": max_index},
D
dzhwinter 已提交
410
        attrs={"pooltype": pool_type.upper()})
D
dzhwinter 已提交
411 412 413 414

    return pool_out


F
fengjiayi 已提交
415 416 417 418 419 420
def pool2d(input,
           pool_size,
           pool_type,
           pool_stride=[1, 1],
           pool_padding=[0, 0],
           global_pooling=False,
421 422
           main_program=None,
           startup_program=None):
F
fengjiayi 已提交
423 424 425 426 427 428 429 430 431 432 433
    if pool_type not in ["max", "avg"]:
        raise ValueError(
            "Unknown pool_type: '%s'. It can only be 'max' or 'avg'.",
            str(pool_type))
    if isinstance(pool_size, int):
        pool_size = [pool_size, pool_size]
    if isinstance(pool_stride, int):
        pool_stride = [pool_stride, pool_stride]
    if isinstance(pool_padding, int):
        pool_padding = [pool_padding, pool_padding]

D
dzhwinter 已提交
434
    helper = LayerHelper('pool2d', **locals())
F
fengjiayi 已提交
435 436 437 438 439 440 441 442
    dtype = helper.input_dtype()
    pool_out = helper.create_tmp_variable(dtype)

    helper.append_op(
        type="pool2d",
        inputs={"X": input},
        outputs={"Out": pool_out},
        attrs={
C
chengduoZH 已提交
443
            "pooling_type": pool_type,
F
fengjiayi 已提交
444
            "ksize": pool_size,
C
chengduoZH 已提交
445
            "global_pooling": global_pooling,
F
fengjiayi 已提交
446 447 448 449 450
            "strides": pool_stride,
            "paddings": pool_padding
        })

    return pool_out
Y
Yu Yang 已提交
451 452


Q
Qiao Longfei 已提交
453 454 455 456
def batch_norm(input,
               act=None,
               is_test=False,
               momentum=0.9,
457
               epsilon=1e-05,
Q
Qiao Longfei 已提交
458 459 460
               param_attr=None,
               bias_attr=None,
               data_layout='NCHW',
461 462
               main_program=None,
               startup_program=None):
Q
Qiao Longfei 已提交
463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478
    helper = LayerHelper('batch_norm', **locals())
    dtype = helper.input_dtype()

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

    param_shape = [channel_num]

    # create parameter
    scale = helper.create_parameter(
479 480 481 482
        attr=helper.param_attr,
        shape=param_shape,
        dtype=dtype,
        initializer=ConstantInitializer(1.0))
Q
Qiao Longfei 已提交
483
    bias = helper.create_parameter(
484 485 486 487 488 489 490 491 492 493 494 495 496 497
        attr=helper.param_attr,
        shape=param_shape,
        dtype=dtype,
        initializer=ConstantInitializer(0.0))

    mean = helper.create_global_variable(
        dtype=input.data_type, shape=param_shape, persistable=True)
    helper.set_variable_initializer(
        var=mean, initializer=ConstantInitializer(0.0))

    variance = helper.create_global_variable(
        dtype=input.data_type, shape=param_shape, persistable=True)
    helper.set_variable_initializer(
        var=variance, initializer=ConstantInitializer(1.0))
Q
Qiao Longfei 已提交
498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531

    # create output
    # mean and mean_out share the same memory
    mean_out = mean
    # variance and variance out share the same memory
    variance_out = variance
    saved_mean = helper.create_tmp_variable(dtype)
    saved_variance = helper.create_tmp_variable(dtype)

    batch_norm_out = helper.create_tmp_variable(dtype)

    helper.append_op(
        type="batch_norm",
        inputs={
            "X": input,
            "Scale": scale,
            "Bias": bias,
            "Mean": mean,
            "Variance": variance
        },
        outputs={
            "Y": batch_norm_out,
            "MeanOut": mean_out,
            "VarianceOut": variance_out,
            "SavedMean": saved_mean,
            "SavedVariance": saved_variance
        },
        attrs={"momentum": momentum,
               "epsilon": epsilon,
               "is_test": is_test})

    return helper.append_activation(batch_norm_out)


Y
Yu Yang 已提交
532 533 534 535 536 537
class BlockGuard(object):
    """
    BlockGuard used to create sub-block in program by using Python `with` 
    keyword.
    """

538 539
    def __init__(self, main_program):
        if not isinstance(main_program, Program):
Y
Yu Yang 已提交
540
            raise TypeError("BlockGuard takes a program")
541
        self.main_program = main_program
Y
Yu Yang 已提交
542 543

    def __enter__(self):
544
        self.main_program.create_block()
Y
Yu Yang 已提交
545 546

    def __exit__(self, exc_type, exc_val, exc_tb):
547
        self.main_program.rollback()
Y
Yu Yang 已提交
548 549 550 551 552 553 554 555 556
        if exc_type is not None:
            return False  # re-raise exception
        return True


class StaticRNNGuard(BlockGuard):
    def __init__(self, rnn):
        if not isinstance(rnn, StaticRNN):
            raise TypeError("StaticRNNGuard takes an StaticRNN")
557
        super(StaticRNNGuard, self).__init__(rnn.helper.main_program)
Y
Yu Yang 已提交
558 559 560 561 562 563 564
        self.rnn = rnn

    def __enter__(self):
        self.rnn.status = StaticRNN.IN_RNN_BLOCK
        return super(StaticRNNGuard, self).__enter__()

    def __exit__(self, exc_type, exc_val, exc_tb):
Y
Yu Yang 已提交
565 566
        if exc_type is not None:
            return False
Y
Yu Yang 已提交
567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592
        self.rnn.status = StaticRNN.AFTER_RNN_BLOCK
        self.rnn.complete_rnn_op()
        return super(StaticRNNGuard, self).__exit__(exc_type, exc_val, exc_tb)


class StaticRNNMemoryLink(object):
    """
    :param init: the initial variable for Memory
    :type init: Variable
    :param pre_mem: the memory variable in previous time step
    :type pre_mem: Variable
    :param mem: the memory variable in current time step
    :type mem: Variable
    """

    def __init__(self, init, pre_mem, mem=None):
        self.init = init
        self.pre_mem = pre_mem
        self.mem = mem


class StaticRNN(object):
    BEFORE_RNN_BLOCK = 0
    IN_RNN_BLOCK = 1
    AFTER_RNN_BLOCK = 2

593 594 595
    def __init__(self, name=None, main_program=None):
        self.helper = LayerHelper(
            "static_rnn", name=name, main_program=main_program)
Y
Yu Yang 已提交
596 597 598 599 600 601 602 603 604 605 606 607 608 609
        self.memories = {}  # memory map, from pre_mem.name --> MemoryLink
        self.inputs = []  # input variable list in current block
        self.outputs = []  # output variable list in parent block
        self.status = StaticRNN.BEFORE_RNN_BLOCK  # status flag.
        # sequence length, since it is a static RNN, sequence length are fixed.
        self.seq_len = None

    def step(self):
        return StaticRNNGuard(self)

    def _assert_in_rnn_block_(self, method):
        if self.status != StaticRNN.IN_RNN_BLOCK:
            raise ValueError("You must invoke {0} in rnn block".format(method))

610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625
    def memory(self,
               init=None,
               shape=None,
               batch_ref=None,
               init_value=0.0,
               init_batch_dim_idx=0,
               ref_batch_dim_idx=1):
        '''
        :param init: boot memory, if not set, a shape, batch_ref must be provided
        :param shape: shape of the boot memory
        :param batch_ref: batch size reference variable
        :param init_value: the init value of boot memory
        :param init_batch_dim_idx: the index of batch size in init's dimension
        :param ref_batch_dim_idx: the index of batch size in batch_ref's dimension
        :return: boot memory
        '''
Y
Yu Yang 已提交
626 627
        self._assert_in_rnn_block_('memory')
        if init is None:
628
            if shape is None or batch_ref is None:
Y
Yu Yang 已提交
629
                raise ValueError(
630
                    "if init is None, memory at least need shape and batch_ref")
Y
Yu Yang 已提交
631 632 633
            parent_block = self.parent_block()
            var_name = unique_name("@".join([self.helper.name, "memory_boot"]))
            boot_var = parent_block.create_var(
634 635 636 637
                name=var_name,
                shape=shape,
                dtype=batch_ref.data_type,
                persistable=False)
Y
Yu Yang 已提交
638 639

            parent_block.append_op(
640 641
                type="fill_constant_batch_size_like",
                inputs={'Input': [batch_ref]},
Y
Yu Yang 已提交
642 643 644
                outputs={'Out': [boot_var]},
                attrs={
                    'value': init_value,
645 646 647 648
                    'shape': boot_var.shape,
                    'data_type': boot_var.data_type,
                    'input_dim_idx': ref_batch_dim_idx,
                    'output_dim_idx': init_batch_dim_idx
Y
Yu Yang 已提交
649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665
                })

            return self.memory(init=boot_var)
        else:
            pre_mem = self.helper.create_variable(
                name=unique_name("@".join([self.helper.name, "mem"])),
                dtype=init.data_type,
                shape=init.shape)
            self.memories[pre_mem.name] = StaticRNNMemoryLink(
                init=init, pre_mem=pre_mem)
            return pre_mem

    def step_input(self, x):
        self._assert_in_rnn_block_('step_input')
        if not isinstance(x, Variable):
            raise TypeError("step input takes a Variable")
        if self.seq_len is None:
Y
Yu Yang 已提交
666 667
            self.seq_len = x.shape[0]
        elif self.seq_len != x.shape[0]:
Y
Yu Yang 已提交
668 669 670 671 672
            raise ValueError("Static RNN only take fix seq_len input")

        ipt = self.helper.create_variable(
            name=x.name,
            dtype=x.data_type,
Y
Yu Yang 已提交
673
            shape=list(x.shape[1:]),
Y
Yu Yang 已提交
674 675 676 677 678 679 680 681 682
            type=x.type)
        self.inputs.append(ipt)
        return ipt

    def step_output(self, o):
        self._assert_in_rnn_block_('step_output')
        if not isinstance(o, Variable):
            raise TypeError("step output takes a Variable")

Y
Yu Yang 已提交
683 684 685 686 687 688 689
        tmp_o = self.helper.create_tmp_variable(dtype=o.data_type)
        self.helper.append_op(
            type='rnn_memory_helper',
            inputs={'X': [o]},
            outputs={'Out': tmp_o},
            attrs={'data_type': o.data_type})

Y
Yu Yang 已提交
690
        out_var = self.parent_block().create_var(
Y
Yu Yang 已提交
691 692 693
            name=tmp_o.name,
            shape=[self.seq_len] + list(tmp_o.shape),
            dtype=tmp_o.data_type)
Y
Yu Yang 已提交
694 695 696 697 698 699 700 701 702 703 704 705 706

        self.outputs.append(out_var)

    def output(self, *outputs):
        for each in outputs:
            self.step_output(each)

    def update_memory(self, mem, var):
        if not isinstance(mem, Variable) or not isinstance(var, Variable):
            raise TypeError("update memory should take variables")
        self.memories[mem.name].mem = var

    def parent_block(self):
707
        prog = self.helper.main_program
Y
Yu Yang 已提交
708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723
        parent_idx = prog.current_block().parent_idx
        assert parent_idx >= 0
        parent_block = prog.block(parent_idx)
        return parent_block

    def __call__(self, *args, **kwargs):
        if self.status != StaticRNN.AFTER_RNN_BLOCK:
            raise ValueError("RNN output can only be retrieved after rnn block")
        if len(self.outputs) == 0:
            raise ValueError("RNN has no output")
        elif len(self.outputs) == 1:
            return self.outputs[0]
        else:
            return self.outputs

    def complete_rnn_op(self):
724 725
        main_program = self.helper.main_program
        rnn_block = main_program.current_block()
Y
Yu Yang 已提交
726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788
        parent_block = self.parent_block()

        local_inputs = set()

        for op in rnn_block.ops:
            assert isinstance(op, Operator)
            for oname in op.output_names:
                for out_var_name in op.output(oname):
                    local_inputs.add(out_var_name)

        for var in self.inputs:
            local_inputs.add(var.name)
        for m in self.memories:
            local_inputs.add(m)

        params = list()
        for op in rnn_block.ops:
            assert isinstance(op, Operator)
            for iname in op.input_names:
                for in_var_name in op.input(iname):
                    if in_var_name not in local_inputs:
                        params.append(in_var_name)

        parameters = [parent_block.var(name) for name in params]

        step_scope = parent_block.create_var(
            type=core.VarDesc.VarType.STEP_SCOPES)

        inlinks = [parent_block.var(i.name) for i in self.inputs]
        outlinks = self.outputs

        boot_memories = []
        pre_memories = []
        memories = []
        for _, mem in self.memories.iteritems():
            boot_memories.append(mem.init)
            pre_memories.append(mem.pre_mem.name)
            mem_var = rnn_block.var(mem.mem.name)
            assert isinstance(mem_var, Variable)
            new_mem = self.helper.create_tmp_variable(dtype=mem_var.data_type)

            rnn_block.append_op(
                type='rnn_memory_helper',
                inputs={'X': [mem_var]},
                outputs={'Out': [new_mem]},
                attrs={'data_type': mem_var.data_type})

            memories.append(new_mem.name)

        parent_block.append_op(
            type='recurrent',
            inputs={
                'inputs': inlinks,
                'initial_states': boot_memories,
                'parameters': parameters
            },
            outputs={'outputs': outlinks,
                     'step_scopes': [step_scope]},
            attrs={
                'ex_states': pre_memories,
                'states': memories,
                'step_block': rnn_block
            })
Y
Yu Yang 已提交
789 790


Y
Yang Yang(Tony) 已提交
791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830
def lstm(x,
         c_pre_init,
         hidden_dim,
         forget_bias=None,
         main_program=None,
         startup_program=None):
    helper = LayerHelper('lstm_unit', **locals())
    rnn = StaticRNN()
    with rnn.step():
        c_pre = rnn.memory(init=c_pre_init)
        x_t = rnn.step_input(x)

        before_fc = concat(
            input=[x_t, c_pre],
            axis=1,
            main_program=main_program,
            startup_program=startup_program)
        after_fc = fc(input=before_fc,
                      size=hidden_dim * 4,
                      main_program=main_program,
                      startup_program=startup_program)

        data_type = x.data_type
        c = helper.create_tmp_variable(data_type)
        h = helper.create_tmp_variable(data_type)

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

        rnn.update_memory(c_pre, c)
        rnn.output(h)

    return rnn()


831
def lod_rank_table(x, level=0, main_program=None):
Y
Yu Yang 已提交
832 833 834 835 836 837 838 839 840 841
    helper = LayerHelper("lod_rank_table", **locals())
    table = helper.create_variable(
        type=core.VarDesc.VarType.LOD_RANK_TABLE,
        name=unique_name("lod_rank_table"))
    helper.append_op(
        type='lod_rank_table',
        inputs={'X': x},
        outputs={'Out': table},
        attrs={'level': level})
    return table
Y
Yu Yang 已提交
842 843


844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867
def lod_tensor_to_array(x, table, main_program=None):
    helper = LayerHelper("lod_tensor_to_array", **locals())
    array = helper.create_variable(
        name=unique_name("lod_tensor_to_array"),
        type=core.VarDesc.VarType.LOD_TENSOR_ARRAY)
    helper.append_op(
        type='lod_tensor_to_array',
        inputs={'X': x,
                'RankTable': table},
        outputs={'Out': array})
    return array


def array_to_lod_tensor(x, table, main_program=None):
    helper = LayerHelper("array_to_lod_tensor", **locals())
    tmp = helper.create_tmp_variable(dtype=x.data_type)
    helper.append_op(
        type="array_to_lod_tensor",
        inputs={'X': x,
                'RankTable': table},
        outputs={'Out': tmp})
    return tmp


Y
Yu Yang 已提交
868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893
def fill_constant(shape, dtype, value, main_program=None):
    helper = LayerHelper("ones", **locals())
    out = helper.create_tmp_variable(dtype=dtype)
    helper.append_op(
        type='fill_constant',
        inputs={},
        outputs={'Out': [out]},
        attrs={
            'shape': shape,
            'data_type': out.data_type,
            'value': float(value)
        })
    out.stop_gradient = True
    return out


def ones(shape, dtype, main_program=None):
    return fill_constant(value=1.0, **locals())


def zeros(shape, dtype, main_program=None):
    return fill_constant(value=0.0, **locals())


def increment(x, value=1.0, main_program=None):
    helper = LayerHelper("increment", **locals())
Y
Yang Yu 已提交
894
    out = helper.create_tmp_variable(dtype=x.data_type)
Y
Yu Yang 已提交
895 896 897
    helper.append_op(
        type='increment',
        inputs={'X': [x]},
Y
Yang Yu 已提交
898
        outputs={'Out': [out]},
Y
Yu Yang 已提交
899
        attrs={'step': value})
Y
Yang Yu 已提交
900
    return out
Y
Yu Yang 已提交
901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930


def array_write(x, i, array=None, main_program=None):
    helper = LayerHelper('array_write', **locals())
    if array is None:
        array = helper.create_variable(
            name="{0}.out".format(helper.name),
            type=core.VarDesc.VarType.LOD_TENSOR_ARRAY,
            dtype=x.data_type)
    helper.append_op(
        type='write_to_array',
        inputs={'X': [x],
                'I': [i]},
        outputs={'Out': [array]})
    return array


def array_read(array, i, main_program=None):
    helper = LayerHelper('array_read', **locals())
    if not isinstance(
            array,
            Variable) or array.type != core.VarDesc.VarType.LOD_TENSOR_ARRAY:
        raise TypeError("array should be tensor array vairable")
    out = helper.create_tmp_variable(dtype=array.data_type)
    helper.append_op(
        type='read_from_array',
        inputs={'X': [array],
                'I': [i]},
        outputs={'Out': [out]})
    return out
Y
Yang Yu 已提交
931 932 933 934 935 936


def shrink_memory(x, i, table, main_program=None):
    helper = LayerHelper('shrink_memory', **locals())
    out = helper.create_tmp_variable(dtype=x.data_type)
    helper.append_op(
Y
Yang Yu 已提交
937
        type='shrink_rnn_memory',
Y
Yang Yu 已提交
938 939 940 941 942 943
        inputs={'X': [x],
                'I': [i],
                'RankTable': [table]},
        outputs={'Out': [out]},
        attrs={})
    return out