framework.py 42.9 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
import collections
Q
qiaolongfei 已提交
16
import contextlib
F
fengjiayi 已提交
17
import re
18

Y
Yu Yang 已提交
19
import numpy as np
Q
qiaolongfei 已提交
20

21
import proto.framework_pb2 as framework_pb2
Q
qiaolongfei 已提交
22
from . import core
Y
Yu Yang 已提交
23
import unique_name
Y
Yu Yang 已提交
24

25
__all__ = [
26 27 28 29 30 31 32 33 34
    'Block',
    'Variable',
    'Program',
    'Operator',
    'default_startup_program',
    'default_main_program',
    'program_guard',
    'switch_startup_program',
    'switch_main_program',
X
xuwei06 已提交
35
    'get_var',
36
]
Y
Yu Yang 已提交
37

Q
qiaolongfei 已提交
38 39 40 41 42 43 44 45 46 47 48 49
EMPTY_VAR_NAME = core.kEmptyVarName()
TEMP_VAR_NAME = core.kTempVarName()
GRAD_VAR_SUFFIX = core.kGradVarSuffix()
ZERO_VAR_SUFFIX = core.kZeroVarSuffix()


def grad_var_name(var_name):
    """
    return gradient name for a certain var name
    """
    return var_name + GRAD_VAR_SUFFIX

Y
Yu Yang 已提交
50

51
def convert_np_dtype_to_dtype_(np_dtype):
52 53 54 55 56
    """
    Convert the data type in numpy to the data type in Paddle
    Args:
        np_dtype(np.dtype): the data type in numpy

57
    Returns(core.VarDesc.VarType): the data type in Paddle
58 59

    """
60 61
    dtype = np.dtype(np_dtype)
    if dtype == np.float32:
62
        return core.VarDesc.VarType.FP32
63
    elif dtype == np.float64:
64
        return core.VarDesc.VarType.FP64
65
    elif dtype == np.float16:
66
        return core.VarDesc.VarType.FP16
67
    elif dtype == np.int32:
68
        return core.VarDesc.VarType.INT32
69
    elif dtype == np.int16:
70
        return core.VarDesc.VarType.INT16
71
    elif dtype == np.int64:
72
        return core.VarDesc.VarType.INT64
73
    elif dtype == np.bool:
74
        return core.VarDesc.VarType.BOOL
75 76 77 78 79
    else:
        raise ValueError("Not supported numpy dtype " + str(dtype))


def dtype_is_floating(dtype):
80 81 82
    """
    Check the data type is floating or not.
    Args:
83
        dtype(np.dtype|core.VarDesc.VarType): data type.
84 85 86 87 88
            Could be numpy format or Paddle format

    Returns(bool): True if data type is a float value

    """
89
    if not isinstance(dtype, core.VarDesc.VarType):
90 91
        dtype = convert_np_dtype_to_dtype_(dtype)

92 93 94 95
    return dtype in [
        core.VarDesc.VarType.FP16, core.VarDesc.VarType.FP32,
        core.VarDesc.VarType.FP64
    ]
96 97


Y
Yang Yang(Tony) 已提交
98
def _debug_string_(proto, throw_on_error=True):
99 100 101 102 103 104 105 106 107 108 109
    """
    Get the debug string of a protobuf message. The message could be not
    initialized.
    Args:
        proto(google.protobuf.message.Message): The protobuf message
        throw_on_error(bool): True if raise an error when the protobuf message
            is not initialized.

    Returns(str): The debug string of the protobuf message

    """
Y
Yu Yang 已提交
110
    error_fields = list()
Y
Yang Yang(Tony) 已提交
111
    if not proto.IsInitialized(error_fields) and throw_on_error:
C
caoying03 已提交
112 113
        raise ValueError("{0} are not initialized.\nThe message is {1}:\n".
                         format(error_fields, proto))
Y
Yu Yang 已提交
114 115 116
    return proto.__str__()


Y
Yu Yang 已提交
117
class Variable(object):
118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140
    """
    Python variable. Every input and output of an operator is a variable. Every
    variable belongs to a block. The variable has a name and two variables in
    different blocks could have the same name.

    There are many kinds of variables. Please reference the framework.proto for
    details.

    Notes: The constructor of Variable should not be invoked directly. Please
    use `Block.create_var` to create a variable.

    >>> cur_program = Program()
    >>> cur_block = cur_program.current_block()
    >>> new_variable = cur_block.create_var(
    >>>                    name="X", shape=[-1, 23, 48], dtype='float32')

    Args:
        block(Block): The associated block. It will be passed by
            `Block.create_var` automatically.
        type(core.VarDesc.VarType): Variable type. Please reference the
            framework.proto for details.
        shape(tuple|list|None): The shape of variable. -1 means the batch size.
            Some kinds of variable do not contain shape, just set it to None.
141
        dtype(np.dtype|core.VarDesc.VarType|str): The data type of variable.
142
        lod_level(int): The level of lod tensor. 0 means it is not a time
143
            series data.
144 145
        capacity(int): The capacity of Channel variable. Ignored
            for other types.
146 147 148 149 150 151
        persistable(bool): True if the variable should be saved as check point.
            Defaults to False.
        stop_gradient(bool): True if the variable will stop to calculate
            gradients when backward. Defaults to False.
    """

Y
Yu Yang 已提交
152 153
    def __init__(self,
                 block,
Y
Yu Yang 已提交
154
                 type=core.VarDesc.VarType.LOD_TENSOR,
Y
Yu Yang 已提交
155 156 157 158
                 name=None,
                 shape=None,
                 dtype=None,
                 lod_level=None,
159
                 capacity=None,
Q
QI JUN 已提交
160
                 persistable=None,
F
fengjiayi 已提交
161
                 error_clip=None,
Y
Yu Yang 已提交
162
                 stop_gradient=False,
Y
Yu Yang 已提交
163
                 **kwargs):
Y
Yu Yang 已提交
164
        self.block = block
F
fengjiayi 已提交
165
        self.error_clip = error_clip
Y
Yu Yang 已提交
166 167

        if name is None:
Y
Yu Yang 已提交
168
            name = unique_name.generate('_generated_var')
D
Dong Zhihong 已提交
169 170 171 172
        is_new_var = False
        self.desc = self.block.desc.find_var(name)

        if self.desc is None:
D
dongzhihong 已提交
173
            self.desc = self.block.desc.var(name)
Y
Yu Yang 已提交
174
            is_new_var = True
Y
Yu Yang 已提交
175

Y
Yu Yang 已提交
176 177 178 179 180 181 182 183
        if is_new_var:
            self.desc.set_type(type)
        elif self.desc.type() != type:
            raise ValueError("Variable {0} has been created before. The "
                             "previous type is {1}; the new type is {2}. They"
                             " are not matched".format(self.name,
                                                       self.desc.type(), type))

Y
Yu Yang 已提交
184
        if shape is not None:
Y
Yu Yang 已提交
185
            if is_new_var:
186
                self.desc.set_shape(shape)
Y
Yu Yang 已提交
187 188 189 190 191 192 193 194
            else:
                old_shape = self.shape
                shape = tuple(shape)
                if shape != old_shape:
                    raise ValueError(
                        "Variable {0} has been created before. the previous "
                        "shape is {1}; the new shape is {2}. They are not "
                        "matched.".format(self.name, old_shape, shape))
Y
Yu Yang 已提交
195
        if dtype is not None:
196
            if not isinstance(dtype, core.VarDesc.VarType):
197
                dtype = convert_np_dtype_to_dtype_(dtype)
Y
Yu Yang 已提交
198
            if is_new_var:
F
fengjiayi 已提交
199
                self.desc.set_dtype(dtype)
Y
Yu Yang 已提交
200
            else:
F
fengjiayi 已提交
201
                old_dtype = self.dtype
Q
QI JUN 已提交
202
                if dtype != old_dtype:
Y
Yu Yang 已提交
203 204 205 206 207
                    raise ValueError("Variable {0} has been created before. "
                                     "The previous data type is {1}; the new "
                                     "data type is {2}. They are not "
                                     "matched.".format(self.name, old_dtype,
                                                       dtype))
Y
Yu Yang 已提交
208 209

        if lod_level is not None:
Y
Yu Yang 已提交
210
            if is_new_var:
211
                self.desc.set_lod_level(lod_level)
Y
Yu Yang 已提交
212 213 214 215 216 217 218
            else:
                if lod_level != self.lod_level:
                    raise ValueError("Variable {0} has been created before. "
                                     "The previous lod_level is {1}; the new "
                                     "lod_level is {2}. They are not "
                                     "matched".format(self.name, self.lod_level,
                                                      lod_level))
219 220 221 222 223 224 225 226 227 228 229
        if persistable is not None:
            if is_new_var:
                self.desc.set_persistable(persistable)
            else:
                if persistable != self.persistable:
                    raise ValueError(
                        "Variable {0} has been created before."
                        "The previous persistable is {1}; the new "
                        "persistable is {2}. They are not matched".format(
                            self.name, self.persistable, persistable))

230 231 232 233 234 235 236 237
        if capacity is not None:
            if is_new_var:
                self.desc.set_capacity(capacity)
            else:
                # TODO(abhinavarora) : Compare with set capacity once,
                # get_capacity is implemented
                pass

Y
Yu Yang 已提交
238
        self.block.vars[name] = self
Y
Yu Yang 已提交
239
        self.op = None
Y
Yu Yang 已提交
240
        self.stop_gradient = stop_gradient
Y
Yu Yang 已提交
241

242
    def __str__(self):
Y
Yang Yang(Tony) 已提交
243 244
        return self.to_string(True)

F
update  
fengjiayi 已提交
245
    def to_string(self, throw_on_error, with_details=False):
246 247 248 249 250 251
        """
        Get debug string.

        Args:
            throw_on_error(bool): True if raise an exception when self is not
                intialized.
F
update  
fengjiayi 已提交
252 253
            with_details(bool): more details about variables and parameters
                (e.g. trainable, optimize_attr, ...) will be printed when with_details is True
254 255 256 257

        Returns(str): The debug string.

        """
F
update  
fengjiayi 已提交
258 259
        assert isinstance(throw_on_error, bool) and isinstance(with_details,
                                                               bool)
260 261
        protostr = self.desc.serialize_to_string()
        proto = framework_pb2.VarDesc.FromString(str(protostr))
F
update  
fengjiayi 已提交
262 263 264 265 266 267 268
        res_str = _debug_string_(proto, throw_on_error)
        if with_details:
            additional_attr = ("error_clip", "stop_gradient")
            for attr_name in additional_attr:
                res_str += "%s: %s\n" % (attr_name,
                                         str(getattr(self, attr_name)))
        return res_str
269 270 271

    __repr__ = __str__

272 273 274
    def set_desc(self, input):
        self.desc = input

275 276 277 278
    @property
    def persistable(self):
        return self.desc.persistable()

Y
Yu Yang 已提交
279 280 281 282
    @persistable.setter
    def persistable(self, p):
        self.desc.set_persistable(p)

Y
Yu Yang 已提交
283 284
    @property
    def name(self):
285
        return self.desc.name()
Y
Yu Yang 已提交
286

T
typhoonzero 已提交
287 288 289 290
    @name.setter
    def name(self, new_name):
        self.desc.set_name(new_name)

Y
Yu Yang 已提交
291 292 293
    @property
    def shape(self):
        # convert to tuple, make it as same as numpy API.
294
        return tuple(self.desc.shape())
Y
Yu Yang 已提交
295 296

    @property
F
fengjiayi 已提交
297 298
    def dtype(self):
        return self.desc.dtype()
Y
Yu Yang 已提交
299 300 301

    @property
    def lod_level(self):
302
        return self.desc.lod_level()
Y
Yu Yang 已提交
303

Y
Yu Yang 已提交
304 305 306 307
    @property
    def type(self):
        return self.desc.type()

308 309 310
    def set_error_clip(self, error_clip):
        self.error_clip = error_clip

Y
Yu Yang 已提交
311

F
fengjiayi 已提交
312 313 314
def get_all_op_protos():
    """
    Get all registered op proto from PaddlePaddle C++ end.
315 316 317

    Returns(list): list of OpProto

F
fengjiayi 已提交
318 319 320 321 322 323 324 325 326 327
    """
    protostrs = core.get_all_op_protos()
    ret_values = []
    for pbstr in protostrs:
        op_proto = framework_pb2.OpProto.FromString(str(pbstr))
        ret_values.append(op_proto)
    return ret_values


class OpProtoHolder(object):
328 329 330 331
    """
    A global variable to hold all OpProtos from C++ as a map
    """

F
fengjiayi 已提交
332 333 334 335 336 337 338 339 340
    @classmethod
    def instance(cls):
        if not hasattr(cls, '_instance'):
            cls._instance = cls()
        return cls._instance

    def __init__(self):
        assert not hasattr(
            self.__class__,
341
            '_instance'), 'Please use `instance()` to get OpProtoHolder object!'
F
fengjiayi 已提交
342 343 344 345 346 347
        op_protos = get_all_op_protos()
        self.op_proto_map = {}
        for proto in op_protos:
            self.op_proto_map[proto.type] = proto

    def get_op_proto(self, type):
348 349 350 351 352 353 354 355
        """
        Get OpProto by a type string.
        Args:
            type(str): The type that operator registered in C++ side.

        Returns(framework_pb2.OpProto): The OpProto

        """
Y
Yu Yang 已提交
356 357
        if type not in self.op_proto_map:
            raise ValueError("Operator \"%s\" has not been registered." % type)
F
fengjiayi 已提交
358 359 360
        return self.op_proto_map[type]


Y
Yu Yang 已提交
361
class Operator(object):
362
    """
363 364
    Python Operator class. The operator represents the build in instructions in a
    Block. Users can use the build in instructions to describe their neural
365 366 367
    network.
    """

Y
Yu Yang 已提交
368 369
    def __init__(self,
                 block,
Y
Yu Yang 已提交
370
                 desc,
Y
Yu Yang 已提交
371 372 373 374
                 type=None,
                 inputs=None,
                 outputs=None,
                 attrs=None):
375 376 377 378 379 380 381 382 383 384 385 386 387 388
        """
        Constructor.

        Notes: The constructor of operator should not be invoked directly. Use
        Block.append_op or Block.prepend_op instead.

        >>> cur_program = Program()
        >>> cur_block = cur_program.current_block()
        >>> # var1 += var2 + var3
        >>> cur_block.append_op(type="sum",
        >>>                     inputs={"X": [var1, var2, var3]},
        >>>                     outputs={"Out": [var1]})

        Args:
C
caoying03 已提交
389 390
            block(Block): The block has the current operator.
            desc(core.OpDesc): The protobuf description.
391 392 393
            type(str): The type of operator.
            inputs(dict): The input dictionary. Key is the input parameter name.
                Value is a list of variables.
C
caoying03 已提交
394 395
            outputs(dict): The output dictionary which has the same format with
                           inputs.
396 397 398 399
            attrs(dict): The attributes dictionary. Key is attribute name. Value
                is the attribute value. The attribute type should be as same as
                the type registered in C++
        """
Y
Yu Yang 已提交
400
        self.block = block
Y
Yu Yang 已提交
401
        self.desc = desc
T
typhoonzero 已提交
402
        self.attrs = attrs
F
fengjiayi 已提交
403 404 405 406 407
        if len(self.desc.type()) != 0:
            return
        if type is None:
            raise ValueError(
                "`type` to initilized an Operator can not be None.")
F
Update  
fengjiayi 已提交
408
        self.desc.set_type(type)
F
fengjiayi 已提交
409
        proto = OpProtoHolder.instance().get_op_proto(type)
410

Y
Yang Yang(Tony) 已提交
411 412
        def find_name(var_list, name):
            for var_name in var_list:
Q
Qiao Longfei 已提交
413
                if var_list[var_name] is not None and var_name == name:
Y
Yang Yang(Tony) 已提交
414 415
                    return True
            return False
Q
QI JUN 已提交
416

Y
Yang Yang(Tony) 已提交
417 418 419 420 421 422 423
        if inputs is not None:
            for in_proto in proto.inputs:
                found = find_name(inputs, in_proto.name)
                assert found or in_proto.dispensable, "Input {} not found".format(
                    in_proto.name)

                if found:
424 425 426 427
                    in_args = inputs[in_proto.name]
                    if not isinstance(in_args, list):
                        in_args = [in_args]
                    if not in_proto.duplicable and len(in_args) > 1:
Y
Yang Yang(Tony) 已提交
428 429
                        raise ValueError(
                            "Input %s expects only one input, but %d are given."
430 431 432
                            % (in_proto.name, len(in_args)))
                    in_arg_names = []
                    for arg in in_args:
Y
Yang Yu 已提交
433 434 435 436
                        if isinstance(arg, basestring):
                            in_arg_names.append(arg)
                        else:
                            in_arg_names.append(arg.name)
437
                    self.desc.set_input(in_proto.name, in_arg_names)
Y
Yang Yang(Tony) 已提交
438 439
                else:
                    self.desc.set_input(in_proto.name, [])
F
Update  
fengjiayi 已提交
440

Y
Yu Yang 已提交
441
        if outputs is not None:
442 443 444 445 446 447 448
            given = set()
            need = set()
            for n in outputs:
                given.add(n)
            for m in proto.outputs:
                need.add(m.name)
            if not given == need:
C
caoying03 已提交
449 450
                raise ValueError(("Incorrect setting for output(s) of "
                                  "operator \"%s\". Need: [%s] Given: [%s]") %
451 452
                                 (type, ", ".join(str(e) for e in need),
                                  ", ".join(str(e) for e in given)))
453

F
fengjiayi 已提交
454
            for out_proto in proto.outputs:
455 456 457 458
                out_args = outputs[out_proto.name]
                if not isinstance(out_args, list):
                    out_args = [out_args]
                if not out_proto.duplicable and len(out_args) > 1:
F
Update  
fengjiayi 已提交
459 460
                    raise ValueError(
                        "Output %s expects only one output, but %d are given." %
461 462 463 464 465 466
                        (out_proto.name, len(out_args)))
                out_arg_names = []
                for arg in out_args:
                    out_arg_names.append(arg.name)
                    arg.op = self
                self.desc.set_output(out_proto.name, out_arg_names)
F
Update  
fengjiayi 已提交
467

Y
Yu Yang 已提交
468
        if attrs is not None:
469 470
            if not isinstance(attrs, dict):
                raise TypeError("'attrs' should be a dict.")
F
fengjiayi 已提交
471
            for attr in proto.attrs:
F
Update  
fengjiayi 已提交
472
                attr_name = attr.name
473
                if (attr_name not in attrs) or (attrs[attr_name] is None):
F
Update  
fengjiayi 已提交
474
                    continue
Y
Yang Yang(Tony) 已提交
475
                if isinstance(attrs[attr_name], Block):
F
Update  
fengjiayi 已提交
476
                    self.desc.set_block_attr(attr_name, attrs[attr_name].desc)
T
typhoonzero 已提交
477 478 479 480
                elif isinstance(attrs[attr_name], core.BlockDesc) or \
                   isinstance(attrs[attr_name], core.ProgramDesc):
                    self.desc.set_serialized_attr(
                        attr_name, attrs[attr_name].serialize_to_string())
Y
Yang Yang(Tony) 已提交
481 482
                else:
                    self.desc.set_attr(attr_name, attrs[attr_name])
Y
Yu Yang 已提交
483

484
        self.desc.check_attrs()
Y
Yang Yang(Tony) 已提交
485
        no_kernel_op_set = {
486
            'feed', 'fetch', 'save', 'load', 'recurrent', 'go',
T
typhoonzero 已提交
487
            'rnn_memory_helper_grad', 'conditional_block', 'while', 'send',
488
            'recv', 'listen_and_serv', 'parallel_do', 'save_combine',
489
            'load_combine', 'ncclInit', 'channel_create', 'channel_close',
T
Thuan Nguyen 已提交
490
            'channel_send', 'channel_recv', 'select'
Y
Yang Yang(Tony) 已提交
491
        }
492
        if type not in no_kernel_op_set:
Q
QI JUN 已提交
493
            self.desc.infer_var_type(self.block.desc)
Y
Yu Yang 已提交
494
            self.desc.infer_shape(self.block.desc)
F
fengjiayi 已提交
495

Y
Yang Yang(Tony) 已提交
496
    def to_string(self, throw_on_error):
497 498 499 500 501 502 503 504 505
        """
        To debug string.
        Args:
            throw_on_error(bool): raise exception when self is not initialized
                when throw_on_error is True

        Returns(str): The debug string.

        """
506 507
        protostr = self.desc.serialize_to_string()
        proto = framework_pb2.OpDesc.FromString(str(protostr))
Y
Yang Yang(Tony) 已提交
508 509 510 511
        return _debug_string_(proto, throw_on_error)

    def __str__(self):
        return self.to_string(True)
512 513 514

    __repr__ = __str__

F
fengjiayi 已提交
515 516 517 518 519
    @property
    def type(self):
        return self.desc.type()

    def input(self, name):
520 521 522 523 524 525 526 527 528
        """
        Get input arguments by the input parameter name
        Args:
            name(str): The input parameter name

        Returns(list): return the list of argument names associated with the
            specific parameter name.

        """
F
fengjiayi 已提交
529 530
        return self.desc.input(name)

T
typhoonzero 已提交
531 532 533 534 535 536
    def rename_input(self, old_name, new_name):
        self.desc.rename_input(old_name, new_name)

    def rename_output(self, old_name, new_name):
        self.desc.rename_output(old_name, new_name)

F
fengjiayi 已提交
537 538
    @property
    def input_names(self):
539 540 541 542 543
        """
        Get all input parameter names
        Returns(list): return a list of input parameter names

        """
F
fengjiayi 已提交
544 545
        return self.desc.input_names()

T
typhoonzero 已提交
546 547 548 549 550 551 552 553
    @property
    def input_arg_names(self):
        return self.desc.input_arg_names()

    @property
    def output_arg_names(self):
        return self.desc.output_arg_names()

F
fengjiayi 已提交
554
    def output(self, name):
555 556 557 558 559 560 561 562 563
        """
        Get output arguments by the output parameter name
        Args:
            name(str): The output parameter name

        Returns(list): return the list of argument names associated with the
            specific parameter name.

        """
F
fengjiayi 已提交
564 565 566 567
        return self.desc.output(name)

    @property
    def output_names(self):
568 569 570 571 572
        """
        Get all output parameter names
        Returns(list): return a list of output parameter names

        """
F
fengjiayi 已提交
573 574
        return self.desc.output_names()

575 576
    @property
    def idx(self):
577 578 579 580 581 582
        """
        Return the array index of current operator.
        Returns(int): The array index in block.ops array
        Raises:
            ValueError: when the operator is not found.
        """
583 584 585 586 587 588
        for i, op in enumerate(self.block.ops):
            if op == self:
                return i
        raise ValueError(
            "Can't find op itself in it's block. It could be a bug of Paddle.")

F
fengjiayi 已提交
589
    def has_attr(self, name):
590 591 592 593 594 595 596 597
        """
        operator has the attribute with name or not.
        Args:
            name(str): the attribute name

        Returns(bool): True if has this attribute.

        """
F
fengjiayi 已提交
598 599 600
        return self.desc.has_attr(name)

    def attr_type(self, name):
601 602 603 604 605 606 607 608
        """
        Get the type of attribute by attribute name
        Args:
            name(str): the attribute name

        Returns(core.AttrType): the attribute type

        """
F
fengjiayi 已提交
609 610 611 612
        return self.desc.attr_type(name)

    @property
    def attr_names(self):
613 614 615 616 617
        """
        Get all attribute names
        Returns(list): The list of attribute name

        """
F
fengjiayi 已提交
618 619 620
        return self.desc.attr_names()

    def attr(self, name):
621 622 623 624 625 626 627 628 629
        """
        Get attribute by name
        Args:
            name(str): the attribute name

        Returns(bool|int|str|float|list): The attribute value. The return value
            can be any valid attribute type.

        """
F
fengjiayi 已提交
630
        return self.desc.attr(name)
Y
Yu Yang 已提交
631

F
fengjiayi 已提交
632
    def block_attr(self, name):
633 634 635 636 637 638 639 640
        """
        Get the block attribute by name
        Args:
            name(str): the attribute name

        Returns(int): the block index

        """
F
fengjiayi 已提交
641
        return self.desc.block_attr(name)
Y
Yu Yang 已提交
642

J
JiayiFeng 已提交
643
    def all_attrs(self):
F
fengjiayi 已提交
644 645 646 647 648 649 650 651 652 653 654 655 656
        """
        Get the attribute dict
        Returns(dict): The Operator's attribute dict
        """
        attr_names = self.attr_names
        attr_map = {}
        for n in attr_names:
            if n == 'sub_block':
                attr_map[n] = self.block_attr(n)
            else:
                attr_map[n] = self.attr(n)
        return attr_map

Y
Yu Yang 已提交
657

Y
Yu Yang 已提交
658 659
class Block(object):
    def __init__(self, program, idx):
Y
Yu Yang 已提交
660
        self.desc = program.desc.block(idx)
Y
Yu Yang 已提交
661
        self.vars = dict()  # var_name --> var
Q
qiaolongfei 已提交
662
        self.ops = list()  # operator list
Y
Yu Yang 已提交
663
        self.program = program
T
typhoonzero 已提交
664
        self.removed_vars = dict()
Y
Yu Yang 已提交
665

666
    def __str__(self):
Y
Yang Yang(Tony) 已提交
667 668
        return self.to_string(True)

F
fengjiayi 已提交
669 670 671 672 673 674
    def to_string(self, throw_on_error, with_details=False):
        """
        To debug string.
        Args:
            throw_on_error(bool): raise exception when self is not initialized
                when throw_on_error is True
F
update  
fengjiayi 已提交
675 676
            with_details(bool): more details about variables and parameters
                (e.g. trainable, optimize_attr, ...) will be printed when with_details is True
F
fengjiayi 已提交
677 678 679 680 681 682 683

        Returns(str): The debug string.

        """
        assert isinstance(throw_on_error, bool) and isinstance(with_details,
                                                               bool)
        if with_details:
F
fengjiayi 已提交
684
            re_add_indent = re.compile(r"\n(.)")
F
fengjiayi 已提交
685 686 687
            res_str = "blocks {\n  idx: %d\n  parent_idx: %d" % (
                self.idx, self.parent_idx)
            for var in self.vars.itervalues():
F
fengjiayi 已提交
688
                res_str += "\n  vars {\n    %s  }" % re_add_indent.sub(
F
update  
fengjiayi 已提交
689
                    r"\n    \1", var.to_string(throw_on_error, with_details))
F
fengjiayi 已提交
690
            for op in self.ops:
F
fengjiayi 已提交
691 692
                res_str += "\n  ops {\n    %s  }" % re_add_indent.sub(
                    r"\n    \1", op.to_string(throw_on_error))
F
fengjiayi 已提交
693 694 695 696 697 698
            res_str += "\n}"
        else:
            protostr = self.desc.serialize_to_string()
            proto = framework_pb2.BlockDesc.FromString(str(protostr))
            res_str = _debug_string_(proto, throw_on_error)
        return res_str
699 700 701

    __repr__ = __str__

Y
Yu Yang 已提交
702 703
    @property
    def parent_idx(self):
Y
Yu Yang 已提交
704
        return self.desc.parent
Y
Yu Yang 已提交
705

Y
Yu Yang 已提交
706 707 708 709 710 711 712
    @property
    def forward_block_idx(self):
        return self.desc.get_forward_block_idx()

    def set_forward_block_idx(self, idx):
        self.desc.set_forward_block_idx(idx)

Y
Yu Yang 已提交
713 714
    @property
    def idx(self):
Y
Yu Yang 已提交
715
        return self.desc.id
Y
Yu Yang 已提交
716

Q
Qiao Longfei 已提交
717
    def var(self, name):
Y
Yu Yang 已提交
718 719 720 721
        if not isinstance(name, basestring):
            raise TypeError()
        v = self.vars.get(name, None)
        if v is None:
Q
Qiao Longfei 已提交
722
            raise ValueError("var %s not in this block" % name)
Y
Yu Yang 已提交
723
        return v
Q
Qiao Longfei 已提交
724

F
fengjiayi 已提交
725
    def var_recursive(self, name):
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
        frontier = list()
        visited = set()

        frontier.append(self)

        prog = self.program

        while len(frontier) != 0:  # BFS
            cur = frontier[0]
            frontier = frontier[1:]

            if id(cur) in visited:
                continue

            if cur.has_var(name):
                return cur.var(name)

            if cur.parent_idx != -1:
                frontier.append(prog.block(cur.parent_idx))

            if cur.forward_block_idx != -1:
                frontier.append(prog.block(cur.forward_block_idx))

            visited.add(id(cur))

        raise ValueError("Var {0} is not found recursively".format(name))
F
fengjiayi 已提交
752

Q
Qiao Longfei 已提交
753
    def all_parameters(self):
754 755 756 757 758
        return list(self.iter_parameters())

    def iter_parameters(self):
        return (item[1] for item in self.vars.iteritems()
                if isinstance(item[1], Parameter))
Q
Qiao Longfei 已提交
759

Y
Yu Yang 已提交
760
    def create_var(self, *args, **kwargs):
761
        var = Variable(block=self, *args, **kwargs)
762 763
        if 'initializer' in kwargs:
            kwargs['initializer'](var, self)
Q
Qiao Longfei 已提交
764
        return var
Y
Yu Yang 已提交
765

Q
Qiao Longfei 已提交
766 767 768
    def has_var(self, name):
        return name in self.vars

T
typhoonzero 已提交
769 770 771 772 773 774
    def rename_var(self, name, new_name):
        """
        Rename variable in vars and ops' inputs and outputs
        """
        if not self.has_var(name):
            raise ValueError("var %s is not in current" % name)
T
wip  
typhoonzero 已提交
775 776
        v = self.var(name)
        if type(v) == Parameter:
T
typhoonzero 已提交
777
            var_type = "Parameter"
T
wip  
typhoonzero 已提交
778 779 780 781 782 783 784
            stop_gradient = v.stop_gradient
            trainable = v.trainable
            optimize_attr = v.optimize_attr
            regularizer = v.regularizer
            gradient_clip_attr = v.gradient_clip_attr
            error_clip = v.error_clip
        elif type(v) == Variable:
T
typhoonzero 已提交
785
            var_type = "Variable"
T
wip  
typhoonzero 已提交
786 787 788 789
            error_clip = v.error_clip
            stop_gradient = v.stop_gradient
        else:
            raise ValueError("unsupported var type: %s", type(v))
T
typhoonzero 已提交
790
        orig_var_type = v.type
T
typhoonzero 已提交
791
        self.desc.rename_var(name, new_name)
T
typhoonzero 已提交
792
        # NOTE: v is destroyed by C++ after calling rename_var.
T
wip  
typhoonzero 已提交
793
        d = self.desc.find_var(new_name)
T
typhoonzero 已提交
794
        if var_type == "Parameter":
T
wip  
typhoonzero 已提交
795 796 797 798
            var = Parameter(
                self,
                d.shape(),
                d.dtype(),
T
typhoonzero 已提交
799
                type=orig_var_type,
T
wip  
typhoonzero 已提交
800 801 802 803 804 805 806
                name=new_name,
                stop_gradient=stop_gradient,
                trainable=trainable,
                optimize_attr=optimize_attr,
                regularizer=regularizer,
                gradient_clip_attr=gradient_clip_attr,
                error_clip=error_clip)
T
typhoonzero 已提交
807
        elif var_type == "Variable":
T
wip  
typhoonzero 已提交
808 809
            var = Variable(
                self,
T
typhoonzero 已提交
810
                type=orig_var_type,
T
wip  
typhoonzero 已提交
811 812 813 814 815 816 817 818
                name=new_name,
                error_clip=error_clip,
                stop_gradient=stop_gradient)

        # rename the python side, sync_with_cpp will only add
        # new vars/ops to python side.
        self.vars[new_name] = var
        del self.vars[name]
T
typhoonzero 已提交
819
        self.sync_with_cpp()
T
typhoonzero 已提交
820

821 822 823 824 825
    def remove_var(self, name):
        self.sync_with_cpp()
        self.desc.remove_var(name)
        del self.vars[name]

Y
Yu Yang 已提交
826 827
    def create_parameter(self, *args, **kwargs):
        global_block = self.program.global_block()
Q
Qiao Longfei 已提交
828
        param = Parameter(global_block, *args, **kwargs)
829 830
        if 'initializer' in kwargs:
            kwargs['initializer'](param, self)
Q
Qiao Longfei 已提交
831
        return param
Y
Yu Yang 已提交
832

Y
Yu Yang 已提交
833
    def append_op(self, *args, **kwargs):
Y
Yu Yang 已提交
834
        op_desc = self.desc.append_op()
835
        op = Operator(block=self, desc=op_desc, *args, **kwargs)
Y
Yu Yang 已提交
836 837 838
        self.ops.append(op)
        return op

Q
qiaolongfei 已提交
839 840 841 842 843 844 845
    def insert_op(self, index, *args, **kwargs):
        self.sync_with_cpp()
        op_desc = self.desc.insert_op(index)
        op = Operator(block=self, desc=op_desc, *args, **kwargs)
        self.ops.insert(index, op)
        return op

846 847 848 849 850
    def remove_op(self, index):
        self.sync_with_cpp()
        self.desc.remove_op(index, index + 1)
        del self.ops[index]

T
typhoonzero 已提交
851 852
    def delete_ops(self, ops):
        # remove from cpp
853
        # FIXME(typhoonzero): remove only the first occurrence.
T
typhoonzero 已提交
854 855 856 857 858
        try:
            start = list(self.ops).index(ops[0])
            end = list(self.ops).index(ops[-1])
        except Exception, e:
            raise e
859

T
typhoonzero 已提交
860
        self.desc.remove_op(start, end + 1)
T
wip  
typhoonzero 已提交
861

Y
Yancey1989 已提交
862
    def slice_ops(self, start, end):
Q
qiaolongfei 已提交
863
        return self.ops[start:end]
Y
Yancey1989 已提交
864

Y
Yu Yang 已提交
865
    def prepend_op(self, *args, **kwargs):
Y
Yu Yang 已提交
866 867
        op_desc = self.desc.prepend_op()
        op = Operator(self, op_desc, *args, **kwargs)
Q
qiaolongfei 已提交
868
        self.ops.insert(0, op)
Y
Yu Yang 已提交
869 870
        return op

Q
Qiao Longfei 已提交
871
    def sync_with_cpp(self):
872
        """
G
gongweibao 已提交
873
        Sync from the desc on the c++ end.
874 875 876

        This method is used to synchronize the c++ desc instance generated by backward.
        """
Q
Qiao Longfei 已提交
877 878 879 880 881
        # sync variables from cpp
        for var in self.desc.all_vars():
            if not self.has_var(var.name()):
                self.create_var(name=var.name(), desc=var, type=var.type())

882 883 884 885 886
        # sync variables removed from c++ end
        for var in self.vars.keys():
            if not self.desc.find_var(var):
                self.vars.pop(var)

Q
Qiao Longfei 已提交
887
        # sync operators from cpp
888 889 890 891
        ops_in_cpp = []
        for op_idx in range(0, self.desc.op_size()):
            ops_in_cpp.append(self.desc.op(op_idx))

Y
Yu Yang 已提交
892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907
        if len(self.ops) != 0:
            first_op_in_python = self.ops[0].desc
            last_op_in_python = self.ops[len(self.ops) - 1].desc
            start_index = None
            end_index = None
            for index in range(len(ops_in_cpp)):
                if first_op_in_python == ops_in_cpp[index]:
                    start_index = index
                if last_op_in_python == ops_in_cpp[index]:
                    end_index = index
            assert start_index is not None
            assert end_index is not None
            assert start_index <= end_index
        else:
            start_index = 0
            end_index = -1
Q
Qiao Longfei 已提交
908 909 910 911 912

        # sync ops append to the head of cpp_ops
        for index in range((start_index - 1 - 1), -1, -1):
            op_desc = ops_in_cpp[index]
            op = Operator(self, op_desc)
Q
qiaolongfei 已提交
913
            self.ops.insert(0, op)
Q
Qiao Longfei 已提交
914 915 916 917 918 919 920

        # sync ops append to the end of cpp_ops
        for index in range((end_index + 1), len(ops_in_cpp)):
            op_desc = ops_in_cpp[index]
            op = Operator(self, op_desc)
            self.ops.append(op)

921 922 923 924 925 926 927 928 929 930 931 932 933
        # sync ops removed from c++ end
        if end_index != -1 and end_index < len(self.ops):
            ops_in_cpp_index = 0
            ops_in_python_index = 0
            while ops_in_python_index < len(
                    self.ops) and ops_in_cpp_index < len(ops_in_cpp):
                if self.ops[ops_in_python_index].desc != ops_in_cpp[
                        ops_in_cpp_index]:
                    del self.ops[ops_in_python_index]
                else:
                    ops_in_cpp_index += 1
                    ops_in_python_index += 1

Q
Qiao Longfei 已提交
934 935 936 937
        assert len(self.ops) == len(ops_in_cpp)
        for index in range(len(self.ops)):
            assert self.ops[index].desc == ops_in_cpp[index]

938 939
    def copy_param_info_from(self, other):
        """
940
        Copy the information of parameters from the other block
941
        Args:
942
            other(Block): the other block
943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965

        Returns:
            None
        """
        if not isinstance(other, Block):
            raise TypeError("copy_param_info_from should be invoked with Block")
        for p in other.iter_parameters():
            assert isinstance(p, Parameter)
            v = self.vars.get(p.name, None)
            if v is None:
                raise ValueError("copy_param_info_from should be invoked with "
                                 "same topology")
            assert isinstance(v, Variable)
            new_p = Parameter(
                block=self,
                shape=v.shape,
                dtype=v.dtype,
                type=v.type,
                lod_level=v.lod_level,
                stop_gradient=p.stop_gradient,
                trainable=p.trainable,
                optimize_attr=p.optimize_attr,
                regularizer=p.regularizer,
F
fengjiayi 已提交
966
                gradient_clip_attr=p.gradient_clip_attr,
F
fengjiayi 已提交
967
                error_clip=p.error_clip,
968 969 970
                name=v.name)
            self.vars[new_p.name] = new_p

971 972 973 974 975 976 977 978 979 980
    def clone_variable(self, var):
        """
        Clone a variable into current block.
        Args:
            var: the variable to be cloned.

        Returns:
            The new  variable cloned from 'var' in current block.
        """
        assert isinstance(var, Variable)
T
update  
typhoonzero 已提交
981 982 983 984 985
        ret_var = None
        # make STEP_SCOPES var can be safely cloned.
        if var.type == core.VarDesc.VarType.STEP_SCOPES:
            ret_var = self.create_var(
                name=var.name, persistable=var.persistable, type=var.type)
T
typhoonzero 已提交
986 987 988 989 990 991 992
        elif var.type == core.VarDesc.VarType.SELECTED_ROWS:
            ret_var = self.create_var(
                name=var.name,
                shape=var.shape,
                dtype=var.dtype,
                type=var.type,
                persistable=True)
T
update  
typhoonzero 已提交
993 994 995 996 997 998 999 1000 1001
        else:
            ret_var = self.create_var(
                name=var.name,
                shape=var.shape,
                dtype=var.dtype,
                type=var.type,
                lod_level=var.lod_level,
                persistable=True)
        return ret_var
1002

Y
Yu Yang 已提交
1003 1004

class Program(object):
1005 1006
    def __init__(self):
        self.desc = core.ProgramDesc()
Y
Yu Yang 已提交
1007 1008
        self.blocks = [Block(self, 0)]
        self.current_block_idx = 0
D
dzhwinter 已提交
1009
        self._seed = 0
Y
Yu Yang 已提交
1010

1011
    def __str__(self):
Y
Yang Yang(Tony) 已提交
1012 1013
        return self.to_string(True)

F
fengjiayi 已提交
1014 1015 1016 1017 1018 1019
    def to_string(self, throw_on_error, with_details=False):
        """
        To debug string.
        Args:
            throw_on_error(bool): raise exception when self is not initialized
                when throw_on_error is True
F
update  
fengjiayi 已提交
1020 1021
            with_details(bool): more details about variables and parameters
                (e.g. trainable, optimize_attr, ...) will be printed when with_details is True
F
fengjiayi 已提交
1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036

        Returns(str): The debug string.

        """
        assert isinstance(throw_on_error, bool) and isinstance(with_details,
                                                               bool)
        if with_details:
            res_str = ""
            for block in self.blocks:
                res_str += block.to_string(throw_on_error, with_details)
        else:
            protostr = self.desc.serialize_to_string()
            proto = framework_pb2.ProgramDesc.FromString(str(protostr))
            res_str = _debug_string_(proto, throw_on_error)
        return res_str
1037

1038 1039 1040
    def get_desc(self):
        return self.desc

1041 1042 1043 1044
    def clone(self, for_test=False):
        """Clone the Program object

        Set for_test to False when we want to clone the program for training.
1045
        Set for_test to True when we want to clone the program for testing.
1046 1047 1048 1049 1050

        Args:
            for_test(bool): Some operators, such as batch_norm and drop_out ops,
                behave differently in training and testing. If for_test is True,
                the is_test attributes in these operators will be set to True for
1051 1052
                testing purposes, otherwise, they remain unchanged.

1053 1054 1055
        Returns(Program):
            The cloned Program object.
        """
Y
Yu Yang 已提交
1056
        p = Program()
1057 1058 1059 1060
        if for_test:
            p.desc = core.inference_optimize(self.desc)
        else:
            p.desc = core.ProgramDesc(self.desc)
Y
Yu Yang 已提交
1061 1062
        p.blocks = [Block(p, i) for i in xrange(self.desc.num_blocks())]
        p.sync_with_cpp()
1063
        p.copy_param_info_from(self)
Y
Yu Yang 已提交
1064
        return p
1065

1066 1067 1068 1069 1070 1071 1072
    def prune(self, targets):
        if not isinstance(targets, list):
            targets = [targets]
        targets_idx = []
        for t in targets:
            if not isinstance(t, Operator):
                if isinstance(t, Variable):
1073 1074 1075 1076 1077 1078
                    if t.op is None:
                        global_block = self.global_block()
                        for op in global_block.ops:
                            if t.name in op.output_arg_names:
                                t.op = op
                                break
1079 1080
                    t = t.op
                else:
C
caoying03 已提交
1081 1082
                    raise ValueError(("All targets of prune() can only be "
                                      "Variable or Operator."))
1083 1084 1085 1086 1087 1088 1089 1090

            targets_idx.append([t.block.idx, t.idx])
        res = Program()
        res.desc = core.prune(self.desc, targets_idx)
        res.blocks = [Block(res, i) for i in xrange(res.desc.num_blocks())]
        res.sync_with_cpp()
        return res

1091 1092 1093 1094 1095 1096 1097
    def inference_optimize(self):
        res = Program()
        res.desc = core.inference_optimize(self.desc)
        res.blocks = [Block(res, i) for i in xrange(res.desc.num_blocks())]
        res.sync_with_cpp()
        return res

1098 1099 1100 1101
    @staticmethod
    def parse_from_string(binary_str):
        p = Program()
        p.desc = core.ProgramDesc(binary_str)
1102
        p.blocks = [Block(p, i) for i in xrange(p.desc.num_blocks())]
1103 1104
        p.sync_with_cpp()
        return p
Y
Yu Yang 已提交
1105

D
dzhwinter 已提交
1106 1107 1108 1109 1110 1111 1112 1113 1114 1115
    @property
    def random_seed(self):
        return self._seed

    @random_seed.setter
    def random_seed(self, seed):
        if not isinstance(seed, int):
            raise ValueError("Seed must be a integer.")
        self._seed = seed

Y
Yu Yang 已提交
1116 1117
    def __repr__(self):
        return str(self)
1118

Y
Yu Yang 已提交
1119 1120 1121
    def global_block(self):
        return self.blocks[0]

Q
Qiao Longfei 已提交
1122 1123 1124
    def block(self, index):
        return self.blocks[index]

Y
Yu Yang 已提交
1125 1126 1127
    def current_block(self):
        return self.blocks[self.current_block_idx]

F
update  
fengjiayi 已提交
1128
    def create_block(self, parent_idx=None):
Y
Yu Yang 已提交
1129
        new_block_idx = len(self.blocks)
F
update  
fengjiayi 已提交
1130 1131 1132
        parent = self.current_block() if parent_idx is None else self.block(
            parent_idx)
        self.desc.append_block(parent.desc)
Y
Yu Yang 已提交
1133 1134 1135 1136 1137 1138 1139
        self.current_block_idx = new_block_idx
        self.blocks.append(Block(self, self.current_block_idx))
        return self.current_block()

    def rollback(self):
        self.current_block_idx = self.current_block().parent_idx

Q
Qiao Longfei 已提交
1140 1141 1142 1143 1144 1145
    def sync_with_cpp(self):
        for block_idx in range(len(self.blocks), self.desc.num_blocks()):
            self.blocks.append(Block(self, block_idx))
        for block in self.blocks:
            block.sync_with_cpp()

1146 1147
    def copy_param_info_from(self, other):
        """
1148
        Copy the information of parameters from other program.
1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163
        Args:
            other(Program): Other program

        Returns:
            None
        """
        if not isinstance(other, Program):
            raise TypeError("copy_param_info_from should be invoked with "
                            "Program")

        if len(self.blocks) != len(other.blocks):
            raise ValueError("copy_param_info_from should be invoked with two "
                             "program, with represent the same topology")
        self.global_block().copy_param_info_from(other.global_block())

1164 1165 1166 1167 1168
    def list_vars(self):
        for each_block in self.blocks:
            for each_var in each_block.vars.itervalues():
                yield each_var

Y
Yu Yang 已提交
1169

Y
Yu Yang 已提交
1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180
class Parameter(Variable):
    def __init__(self, block, shape, dtype, **kwargs):
        if shape is None or dtype is None:
            raise ValueError("Parameter must set shape and dtype")
        if len(shape) == 0:
            raise ValueError("Parameter shape cannot be empty")

        for each in shape:
            if each < 0:
                raise ValueError("Parameter shape should not be related with "
                                 "batch-size")
1181 1182 1183

        Variable.__init__(
            self, block, persistable=True, shape=shape, dtype=dtype, **kwargs)
Y
Yu Yang 已提交
1184 1185 1186 1187
        self.trainable = kwargs.get('trainable', True)

        self.optimize_attr = kwargs.get('optimize_attr', {'learning_rate': 1.0})

1188 1189
        self.regularizer = kwargs.get('regularizer', None)

F
fengjiayi 已提交
1190
        self.gradient_clip_attr = kwargs.get('gradient_clip_attr', None)
Y
Yu Yang 已提交
1191

W
wanghaoshuang 已提交
1192
        self.do_model_average = kwargs.get('do_model_average', None)
W
wanghaoshuang 已提交
1193

F
fengjiayi 已提交
1194 1195 1196
    def __str__(self):
        return self.to_string(True)

F
update  
fengjiayi 已提交
1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213
    def to_string(self, throw_on_error, with_details=False):
        """
        To debug string.
        Args:
            throw_on_error(bool): raise exception when self is not initialized
                when throw_on_error is True
            with_details(bool): more details about variables and parameters
                (e.g. trainable, optimize_attr, ...) will be printed when with_details is True

        Returns(str): The debug string.

        """
        assert isinstance(throw_on_error, bool) and isinstance(with_details,
                                                               bool)
        if with_details:
            res_str = Variable.to_string(self, throw_on_error, True)
            additional_attr = ("trainable", "optimize_attr", "regularizer",
W
wanghaoshuang 已提交
1214
                               "gradient_clip_attr", "do_model_average")
F
update  
fengjiayi 已提交
1215 1216 1217 1218 1219
            for attr_name in additional_attr:
                res_str += "%s: %s\n" % (attr_name,
                                         str(getattr(self, attr_name)))
        else:
            res_str = Variable.to_string(self, throw_on_error, False)
F
fengjiayi 已提交
1220 1221 1222 1223
        return res_str

    __repr__ = __str__

Y
Yu Yang 已提交
1224

Y
Yu Yang 已提交
1225
# program is a global instance.
Y
Yu Yang 已提交
1226 1227
_main_program_ = Program()
_startup_program_ = Program()
1228

1229

1230
def default_startup_program():
Y
Yu Yang 已提交
1231 1232 1233
    """
    Get default startup program. In startup program, Paddle will initialize
    parameters, initialize nccl handle, etc.
1234

Y
Yu Yang 已提交
1235 1236 1237
    Returns:
        Program: startup program
    """
Y
Yu Yang 已提交
1238
    return _startup_program_
1239

1240

1241
def default_main_program():
Y
Yu Yang 已提交
1242 1243
    """
    Get default main program. The main program is used for training or testing.
1244

Y
Yu Yang 已提交
1245 1246 1247
    Returns:
        Program: main program
    """
Y
Yu Yang 已提交
1248
    return _main_program_
Y
Yu Yang 已提交
1249 1250 1251 1252 1253


def switch_main_program(program):
    """
    Switch the main program to a new program.
1254

Y
Yu Yang 已提交
1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268
    Args:
        program(Program): The new main program

    Returns:
        Program: The previous main program
    """
    global _main_program_
    prev_program = _main_program_
    _main_program_ = program
    return prev_program


def switch_startup_program(program):
    """
1269
    Switch the startup program to a new program
Y
Yu Yang 已提交
1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285
    Args:
        program(Program): The new startup program

    Returns:
        Program: The previous startup program
    """
    global _startup_program_
    prev_program = _startup_program_
    _startup_program_ = program
    return prev_program


@contextlib.contextmanager
def program_guard(main_program, startup_program=None):
    """
    Switch program with `with` statement
1286

Y
Yu Yang 已提交
1287 1288 1289 1290
    Examples:
        >>> with program_guard(Program()):
        >>>   data = fluid.layers.data(...)
        >>>   hidden = fluid.layers.fc(...)
1291

Y
Yu Yang 已提交
1292 1293
    Args:
        main_program(Program): New main program inside `with` statement
1294
        startup_program(Program): New startup program inside `with` statement.
Y
Yu Yang 已提交
1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310
            None means do not change startup program.

    Returns:
        None
    """
    if not isinstance(main_program, Program):
        raise TypeError("main_program should be Program")
    main_program = switch_main_program(main_program)
    if startup_program is not None:
        if not isinstance(startup_program, Program):
            raise TypeError("startup_program should be Program")
        startup_program = switch_startup_program(startup_program)
    yield
    switch_main_program(main_program)
    if startup_program is not None:
        switch_startup_program(startup_program)
X
xuwei06 已提交
1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326


def get_var(name, program=None):
    """
    Get a variable by name from the global block of a program
    Args:
        name(str): name of the variable
        program(Program|None): program object.
             If None, default_global_program() will be used.

    Returns:
        Variable
    """
    if program is None:
        program = default_main_program()
    assert isinstance(name, str)
1327
    assert isinstance(program, Program)
X
xuwei06 已提交
1328 1329

    return program.global_block().var(name)