framework.py 65.1 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
import six
19

Y
Yu Yang 已提交
20
import numpy as np
Q
qiaolongfei 已提交
21

22
from .proto import framework_pb2
23 24
try:
    from . import core
25
except ImportError as e:
26 27 28 29
    raise ImportError(
        """NOTE: You may need to run \"export LD_LIBRARY_PATH=/usr/local/lib:$LD_LIBRARY_PATH\"
    if you encounters \"libmkldnn.so not found\" errors. If you have python
    installed in other directory, replace \"/usr/local/lib\" with your own
Y
yuyang18 已提交
30
    directory. The original error is: \n""" + e.message)
31
except Exception as e:
32
    raise e
33
from . import unique_name
Y
Yu Yang 已提交
34

35
__all__ = [
36 37
    'Program',
    'Operator',
F
fengjiayi 已提交
38
    'Parameter',
39 40 41
    'default_startup_program',
    'default_main_program',
    'program_guard',
X
xuwei06 已提交
42
    'get_var',
43
]
Y
Yu Yang 已提交
44

Q
qiaolongfei 已提交
45 46 47 48 49 50 51 52
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):
    """
53 54
    Returns:
        str: gradient name for a certain var name
Q
qiaolongfei 已提交
55 56 57
    """
    return var_name + GRAD_VAR_SUFFIX

Y
Yu Yang 已提交
58

59
def convert_np_dtype_to_dtype_(np_dtype):
60 61
    """
    Convert the data type in numpy to the data type in Paddle
62

63
    Args:
64
        np_dtype(np.dtype): the data type in numpy.
65

66 67
    Returns:
        core.VarDesc.VarType: the data type in Paddle.
68 69

    """
70 71
    dtype = np.dtype(np_dtype)
    if dtype == np.float32:
72
        return core.VarDesc.VarType.FP32
73
    elif dtype == np.float64:
74
        return core.VarDesc.VarType.FP64
75
    elif dtype == np.float16:
76
        return core.VarDesc.VarType.FP16
77
    elif dtype == np.int32:
78
        return core.VarDesc.VarType.INT32
79
    elif dtype == np.int16:
80
        return core.VarDesc.VarType.INT16
81
    elif dtype == np.int64:
82
        return core.VarDesc.VarType.INT64
83
    elif dtype == np.bool:
84
        return core.VarDesc.VarType.BOOL
85 86
    elif dtype == np.uint16:
        return core.VarDesc.VarType.INT16
87 88
    elif dtype == np.uint8:
        return core.VarDesc.VarType.UINT8
89
    else:
90
        raise ValueError("Not supported numpy dtype " + six.binary_type(dtype))
91 92 93


def dtype_is_floating(dtype):
94 95 96
    """
    Check the data type is floating or not.
    Args:
97
        dtype(np.dtype|core.VarDesc.VarType): data type.
98 99 100 101 102
            Could be numpy format or Paddle format

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

    """
103
    if not isinstance(dtype, core.VarDesc.VarType):
104 105
        dtype = convert_np_dtype_to_dtype_(dtype)

106 107 108 109
    return dtype in [
        core.VarDesc.VarType.FP16, core.VarDesc.VarType.FP32,
        core.VarDesc.VarType.FP64
    ]
110 111


Y
Yang Yang(Tony) 已提交
112
def _debug_string_(proto, throw_on_error=True):
113 114 115 116 117 118 119 120 121 122 123
    """
    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
124
    error_fields = list()
Y
Yang Yang(Tony) 已提交
125
    if not proto.IsInitialized(error_fields) and throw_on_error:
C
caoying03 已提交
126 127
        raise ValueError("{0} are not initialized.\nThe message is {1}:\n".
                         format(error_fields, proto))
Y
Yu Yang 已提交
128 129 130
    return proto.__str__()


Y
Yu Yang 已提交
131
class Variable(object):
132
    """
133 134 135
    In Fluid, every input and output of an operator is a variable. In most
    cases, variables are used for holding different kinds of data or training
    labels. A variable belongs to a block. All variable has its own name and
136
    two variables in different blocks could have the same name.
137

138 139
    There are many kinds of variables. Each kind of them has its own attributes
    and usages. Please reference the framework.proto for details.
140

141
    Most of a Variable's member variables can be setted to be None. It mean
142
    it is not available or will be specified later.
143 144

    Args:
145
        block(Block): The block that the variable belongs to.
146 147
        type(core.VarDesc.VarType): Variable type. Please reference the
            framework.proto for details.
148 149
        name(str|None): The name of the variable. If setted None, it will be
            generated automatically. Default: None
150
        shape(tuple|list|None): The shape of the variable. -1 means the batch size.
151
            Some kinds of variable do not contain shape, just set it to None.
152 153 154
            Default: None
        dtype(np.dtype|core.VarDesc.VarType|str|None): The data type of variable.
            Default: None
155
        lod_level (int|None): The level of lod tensor. 0 means it is not a time
156
            series data.
157
            Default: None
158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179
        capacity (int|None): The capacity of Channel variable. Ignored for other
            types. Default: None
        persistable (bool|None): True if the variable is persistable. A persistable
            variable will not be deleted after an iteration ending. Defaults: None.
        error_clip (BaseErrorClipAttr|None): The error clip attributes of the
            corresponding gradient variable. Default: None
        stop_gradient (bool): True if the variable will stop to calculate its
            gradients when backward. Default: False.
        is_data (bool): True if the variable is an input data. Default: False

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

    Examples:
        .. code-block:: python

            cur_program = Program()
            cur_block = cur_program.current_block()
            new_variable = cur_block.create_var(name="X",
                                                shape=[-1, 23, 48],
                                                dtype='float32')
180 181
    """

Y
Yu Yang 已提交
182 183
    def __init__(self,
                 block,
Y
Yu Yang 已提交
184
                 type=core.VarDesc.VarType.LOD_TENSOR,
Y
Yu Yang 已提交
185 186 187 188
                 name=None,
                 shape=None,
                 dtype=None,
                 lod_level=None,
189
                 capacity=None,
Q
QI JUN 已提交
190
                 persistable=None,
F
fengjiayi 已提交
191
                 error_clip=None,
Y
Yu Yang 已提交
192
                 stop_gradient=False,
F
fengjiayi 已提交
193
                 is_data=False,
Y
Yu Yang 已提交
194
                 **kwargs):
Y
Yu Yang 已提交
195
        self.block = block
F
fengjiayi 已提交
196
        self.error_clip = error_clip
Y
Yu Yang 已提交
197 198

        if name is None:
Y
Yu Yang 已提交
199
            name = unique_name.generate('_generated_var')
D
Dong Zhihong 已提交
200
        is_new_var = False
201
        name = name if isinstance(name, six.binary_type) else name.encode()
D
Dong Zhihong 已提交
202 203 204
        self.desc = self.block.desc.find_var(name)

        if self.desc is None:
D
dongzhihong 已提交
205
            self.desc = self.block.desc.var(name)
Y
Yu Yang 已提交
206
            is_new_var = True
Y
Yu Yang 已提交
207

Y
Yu Yang 已提交
208 209 210 211 212 213 214 215
        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
216
        if shape is not None:
Y
Yu Yang 已提交
217
            if is_new_var:
218
                self.desc.set_shape(shape)
Y
Yu Yang 已提交
219 220 221 222 223 224 225 226
            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
227
        if dtype is not None:
228
            if not isinstance(dtype, core.VarDesc.VarType):
229
                dtype = convert_np_dtype_to_dtype_(dtype)
Y
Yu Yang 已提交
230
            if is_new_var:
F
fengjiayi 已提交
231
                self.desc.set_dtype(dtype)
Y
Yu Yang 已提交
232
            else:
F
fengjiayi 已提交
233
                old_dtype = self.dtype
Q
QI JUN 已提交
234
                if dtype != old_dtype:
Y
Yu Yang 已提交
235 236 237 238 239
                    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
240 241

        if lod_level is not None:
Y
Yu Yang 已提交
242
            if is_new_var:
243
                self.desc.set_lod_level(lod_level)
Y
Yu Yang 已提交
244 245 246 247 248 249 250
            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))
251 252 253 254 255 256 257 258 259 260 261
        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))

262 263 264 265 266 267 268 269
        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
270
        self.block.vars[name] = self
Y
Yu Yang 已提交
271
        self.op = None
Y
Yu Yang 已提交
272
        self.stop_gradient = stop_gradient
F
fengjiayi 已提交
273
        self.is_data = is_data
Y
Yu Yang 已提交
274

275
    def __str__(self):
Y
Yang Yang(Tony) 已提交
276 277
        return self.to_string(True)

F
update  
fengjiayi 已提交
278
    def to_string(self, throw_on_error, with_details=False):
279 280 281 282
        """
        Get debug string.

        Args:
283 284
            throw_on_error(bool): True if raise an exception when self is
                not initialized.
F
update  
fengjiayi 已提交
285
            with_details(bool): more details about variables and parameters
286 287
                (e.g. trainable, optimize_attr, ...) will be printed when
                with_details is True. Default False;
288

289 290
        Returns:
            str: The debug string.
291
        """
F
update  
fengjiayi 已提交
292 293
        assert isinstance(throw_on_error, bool) and isinstance(with_details,
                                                               bool)
294
        protostr = self.desc.serialize_to_string()
295
        proto = framework_pb2.VarDesc.FromString(six.binary_type(protostr))
F
update  
fengjiayi 已提交
296 297 298 299
        res_str = _debug_string_(proto, throw_on_error)
        if with_details:
            additional_attr = ("error_clip", "stop_gradient")
            for attr_name in additional_attr:
300 301
                res_str += "%s: %s\n" % (
                    attr_name, six.binary_type(getattr(self, attr_name)))
F
update  
fengjiayi 已提交
302
        return res_str
303 304 305

    __repr__ = __str__

W
Wu Yi 已提交
306
    def _set_desc(self, input):
307 308 309 310 311 312 313 314 315
        """
        Set the variable description.

        Args:
            input(core.VarDesc): The new VarDesc.

        Returns:
            None
        """
316 317
        self.desc = input

318 319 320 321
    @property
    def persistable(self):
        return self.desc.persistable()

Y
Yu Yang 已提交
322 323 324 325
    @persistable.setter
    def persistable(self, p):
        self.desc.set_persistable(p)

Y
Yu Yang 已提交
326 327
    @property
    def name(self):
328
        return self.desc.name()
Y
Yu Yang 已提交
329

T
typhoonzero 已提交
330 331 332 333
    @name.setter
    def name(self, new_name):
        self.desc.set_name(new_name)

Y
Yu Yang 已提交
334 335 336
    @property
    def shape(self):
        # convert to tuple, make it as same as numpy API.
337
        return tuple(self.desc.shape())
Y
Yu Yang 已提交
338 339

    @property
F
fengjiayi 已提交
340 341
    def dtype(self):
        return self.desc.dtype()
Y
Yu Yang 已提交
342 343 344

    @property
    def lod_level(self):
345
        return self.desc.lod_level()
Y
Yu Yang 已提交
346

Y
Yu Yang 已提交
347 348 349 350
    @property
    def type(self):
        return self.desc.type()

W
Wu Yi 已提交
351
    def _set_error_clip(self, error_clip):
352 353 354 355 356 357 358 359 360
        """
        Set the error_clip.

        Args:
            error_clip(BaseErrorClipAttr) : The new error_clip.

        Returns:
            None
        """
361 362
        self.error_clip = error_clip

Y
Yu Yang 已提交
363

F
fengjiayi 已提交
364 365 366
def get_all_op_protos():
    """
    Get all registered op proto from PaddlePaddle C++ end.
367

368 369
    Returns:
       list: list of OpProto.
F
fengjiayi 已提交
370 371 372 373
    """
    protostrs = core.get_all_op_protos()
    ret_values = []
    for pbstr in protostrs:
374
        op_proto = framework_pb2.OpProto.FromString(six.binary_type(pbstr))
F
fengjiayi 已提交
375 376 377 378 379
        ret_values.append(op_proto)
    return ret_values


class OpProtoHolder(object):
380 381 382 383
    """
    A global variable to hold all OpProtos from C++ as a map
    """

F
fengjiayi 已提交
384 385 386 387 388 389 390 391 392
    @classmethod
    def instance(cls):
        if not hasattr(cls, '_instance'):
            cls._instance = cls()
        return cls._instance

    def __init__(self):
        assert not hasattr(
            self.__class__,
393
            '_instance'), 'Please use `instance()` to get OpProtoHolder object!'
F
fengjiayi 已提交
394 395 396 397 398 399
        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):
400 401 402 403 404 405 406 407
        """
        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
408 409
        if type not in self.op_proto_map:
            raise ValueError("Operator \"%s\" has not been registered." % type)
F
fengjiayi 已提交
410 411
        return self.op_proto_map[type]

412 413 414 415 416 417 418
    @staticmethod
    def generated_op_attr_names():
        return {
            core.op_proto_and_checker_maker.kOpRoleAttrName(),
            core.op_proto_and_checker_maker.kOpRoleVarAttrName()
        }

F
fengjiayi 已提交
419

Y
Yu Yang 已提交
420
class Operator(object):
421
    """
422 423 424 425 426 427 428
    In Fluid, all the operation are represented by Operator, and Operator
    is regarded as a build in an instruction of a Block. Users can use the
    build in instructions to describe their neural network.

    Args:
        block(Block): The block has the current operator.
        desc(core.OpDesc): The protobuf description of Operator.
C
chengduoZH 已提交
429
        type(str): The type of operator. Default None.
430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449
        inputs(dict): The input of this Operator. it is a dictionary, for every
            element, key is the input parameter name, and value is a list of
            variables. Default None.
        outputs(dict): The output of this Operator. it is a dictionary, for
            every element, key is the input parameter name, and value is a list
            of variables. Default None.
        attrs(dict): The attributes of this Operator. it is a dictionary, for
            every element, key is attribute name, and value is the attribute value.
            The attribute type should be as same as the type registered in C++ side.
            Default None.

    Returns:
        Operator: The initialized Operator.

    Raises:
        ValueError: If the passed input, output and attrs doesn't match the
            initializing Operator's that registered in C++ side.

    Notes:
        The constructor of operator should not be invoked directly. Use
W
Wu Yi 已提交
450
        Block.append_op or Block._prepend_op instead.
451 452 453 454 455 456 457 458 459 460

    Examples:
        .. code-block:: python

            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]})
461
    """
462 463 464 465 466
    OP_WITHOUT_KERNEL_SET = {
        'feed', 'fetch', 'save', 'load', 'recurrent', 'go',
        'rnn_memory_helper_grad', 'conditional_block', 'while', 'send', 'recv',
        'listen_and_serv', 'parallel_do', 'save_combine', 'load_combine',
        'ncclInit', 'channel_create', 'channel_close', 'channel_send',
T
tangwei12 已提交
467
        'channel_recv', 'select', 'checkpoint_notify', 'gen_nccl_id'
468
    }
469

Y
Yu Yang 已提交
470 471
    def __init__(self,
                 block,
Y
Yu Yang 已提交
472
                 desc,
Y
Yu Yang 已提交
473 474 475 476 477
                 type=None,
                 inputs=None,
                 outputs=None,
                 attrs=None):
        self.block = block
Y
Yu Yang 已提交
478
        self.desc = desc
T
typhoonzero 已提交
479
        self.attrs = attrs
Y
yuyang18 已提交
480 481 482 483 484 485 486 487
        if self.attrs is None:
            self.attrs = dict()
        del attrs

        op_maker = core.op_proto_and_checker_maker

        if op_maker.kOpRoleAttrName() not in self.attrs:
            self.attrs[op_maker.kOpRoleAttrName()] = self.block.program.op_role
Y
yuyang18 已提交
488 489 490 491 492 493 494 495

        role_var_name = op_maker.kOpRoleVarAttrName()
        if len(self.block.program.
               op_role_var) != 0 and role_var_name not in self.attrs:
            self.attrs[role_var_name] = self.block.program.op_role_var

        if role_var_name in self.attrs and len(self.attrs[role_var_name]) == 0:
            del self.attrs[role_var_name]
Y
yuyang18 已提交
496

F
fengjiayi 已提交
497 498 499 500 501
        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
502
        self.desc.set_type(type)
F
fengjiayi 已提交
503
        proto = OpProtoHolder.instance().get_op_proto(type)
504

Y
Yang Yang(Tony) 已提交
505 506
        def find_name(var_list, name):
            for var_name in var_list:
Q
Qiao Longfei 已提交
507
                if var_list[var_name] is not None and var_name == name:
Y
Yang Yang(Tony) 已提交
508 509
                    return True
            return False
Q
QI JUN 已提交
510

Y
Yang Yang(Tony) 已提交
511 512 513 514 515 516 517
        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:
518 519 520 521
                    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
522 523
                        raise ValueError(
                            "Input %s expects only one input, but %d are given."
524 525 526
                            % (in_proto.name, len(in_args)))
                    in_arg_names = []
                    for arg in in_args:
527
                        if issubclass(arg.__class__, six.string_types):
Y
Yang Yu 已提交
528
                            in_arg_names.append(arg)
529 530
                        elif isinstance(arg, six.binary_type):
                            in_arg_names.append(arg.decode())
Y
Yang Yu 已提交
531
                        else:
532 533 534 535
                            if issubclass(arg.name.__class__, six.string_types):
                                in_arg_names.append(arg.name)
                            elif isinstance(arg.name, six.binary_type):
                                in_arg_names.append(arg.name.decode())
536
                    self.desc.set_input(in_proto.name, in_arg_names)
Y
Yang Yang(Tony) 已提交
537 538
                else:
                    self.desc.set_input(in_proto.name, [])
F
Update  
fengjiayi 已提交
539

Y
Yu Yang 已提交
540
        if outputs is not None:
541 542 543 544 545 546 547
            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
548 549
                raise ValueError(("Incorrect setting for output(s) of "
                                  "operator \"%s\". Need: [%s] Given: [%s]") %
550 551 552
                                 (type,
                                  ", ".join(six.binary_type(e) for e in need),
                                  ", ".join(six.binary_type(e) for e in given)))
553

F
fengjiayi 已提交
554
            for out_proto in proto.outputs:
555 556 557 558
                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
559 560
                    raise ValueError(
                        "Output %s expects only one output, but %d are given." %
561 562 563
                        (out_proto.name, len(out_args)))
                out_arg_names = []
                for arg in out_args:
564 565 566 567 568 569
                    if issubclass(arg.name.__class__, six.string_types):
                        out_arg_names.append(arg.name)
                    elif isinstance(arg.name, six.binary_type):
                        out_arg_names.append(arg.name.decode())
                    else:
                        out_arg_names.append(six.u(arg.name))
570 571
                    arg.op = self
                self.desc.set_output(out_proto.name, out_arg_names)
F
Update  
fengjiayi 已提交
572

Y
yuyang18 已提交
573 574
        if self.attrs is not None:
            if not isinstance(self.attrs, dict):
575
                raise TypeError("'attrs' should be a dict.")
F
fengjiayi 已提交
576
            for attr in proto.attrs:
F
Update  
fengjiayi 已提交
577
                attr_name = attr.name
Y
yuyang18 已提交
578 579
                if (attr_name not in self.attrs) or (
                        self.attrs[attr_name] is None):
F
Update  
fengjiayi 已提交
580
                    continue
Y
Yancey1989 已提交
581
                attr_val = self.attrs[attr_name]
G
gongweibao 已提交
582 583
                self._update_desc_attr(attr_name, attr_val)

584
        self.desc.check_attrs()
585
        if self.has_kernel(type):
Q
QI JUN 已提交
586
            self.desc.infer_var_type(self.block.desc)
Y
Yu Yang 已提交
587
            self.desc.infer_shape(self.block.desc)
F
fengjiayi 已提交
588

589 590 591
    def has_kernel(self, op_type):
        return op_type not in self.OP_WITHOUT_KERNEL_SET

Y
Yang Yang(Tony) 已提交
592
    def to_string(self, throw_on_error):
593
        """
594 595
        Get debug string.

596
        Args:
597 598
            throw_on_error(bool): Whether to raise exception if self is not
                initialized.
599

600 601
        Returns:
            str: The debug string.
602 603

        """
604
        protostr = self.desc.serialize_to_string()
605
        proto = framework_pb2.OpDesc.FromString(six.binary_type(protostr))
Y
Yang Yang(Tony) 已提交
606 607 608 609
        return _debug_string_(proto, throw_on_error)

    def __str__(self):
        return self.to_string(True)
610 611 612

    __repr__ = __str__

F
fengjiayi 已提交
613 614 615 616 617
    @property
    def type(self):
        return self.desc.type()

    def input(self, name):
618
        """
619
        Get the input arguments according to the input parameter name.
620

621 622
        Args:
            name(str): The input parameter name.
623

624 625 626
        Returns:
            list: return the list of argument names that associated with \
                the specific parameter name.
627
        """
F
fengjiayi 已提交
628 629
        return self.desc.input(name)

T
typhoonzero 已提交
630
    def rename_input(self, old_name, new_name):
631 632 633 634 635 636 637 638 639 640
        """
        Rename the `old_name` to `new_name`.

        Args:
            old_name(str): The old name of the Operator's input.
            new_name(str): The new name of the Operator's input.

        Returns:
            None
        """
T
typhoonzero 已提交
641 642 643
        self.desc.rename_input(old_name, new_name)

    def rename_output(self, old_name, new_name):
644 645 646 647 648 649 650 651 652 653
        """
        Rename the `old_name` to `new_name`.

        Args:
            old_name(str): The old name of the Operator's output.
            new_name(str): The new name of the Operator's output.

        Returns:
            None
        """
T
typhoonzero 已提交
654 655
        self.desc.rename_output(old_name, new_name)

F
fengjiayi 已提交
656 657 658 659
    @property
    def input_names(self):
        return self.desc.input_names()

T
typhoonzero 已提交
660 661 662 663 664 665 666 667
    @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
668
    def output(self, name):
669
        """
670
        Get output arguments by the output parameter name.
671

672 673
        Args:
            name(str): The output parameter name.
674

675 676 677
        Returns:
            list: return the list of argument names associated with \
                the specific parameter name.
678
        """
F
fengjiayi 已提交
679 680 681 682 683 684
        return self.desc.output(name)

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

685 686 687 688 689 690 691 692
    @property
    def idx(self):
        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
693
    def has_attr(self, name):
694
        """
695 696
        Whether this Operator has the attribute with name or not.

697
        Args:
698
            name(str): the attribute name.
699

700 701
        Returns:
            bool: True if has this attribute.
702 703

        """
F
fengjiayi 已提交
704 705 706
        return self.desc.has_attr(name)

    def attr_type(self, name):
707
        """
708
        Get the type of attribute by attribute's name.
709

710 711
        Args:
            name(str): the attribute name.
712

713 714
        Returns:
            core.AttrType: the attribute type.
715
        """
F
fengjiayi 已提交
716 717
        return self.desc.attr_type(name)

Y
yuyang18 已提交
718
    def set_attr(self, name, val):
719 720 721 722 723 724 725 726 727 728
        """
        Set the value of attribute by attribute's name.

        Args:
            name(str): the attribute name.
            val(bool|int|str|float|list): the value of the attribute.

        Raises:
            ValueError: If the type of value doesn't match with desc.attr_type(name).
        """
Y
yuyang18 已提交
729
        self.attrs[name] = val
G
gongweibao 已提交
730 731 732 733 734 735 736 737 738 739 740 741 742
        self._update_desc_attr(name, val)

    def _update_desc_attr(self, name, val):
        """
        Update the value of desc's attribute by attribute's name.

        Args:
            name(str): the attribute name.
            val(bool|int|str|float|list): the value of the attribute.

        Raises:
            ValueError: If the type of value doesn't match with desc.attr_type(name).
        """
Q
Qiyang Min 已提交
743 744
        if isinstance(val, Block):
            self.desc.set_block_attr(name, val.desc)
Y
Yancey1989 已提交
745 746
        elif isinstance(val, list) and val and all(
                isinstance(v, Block) for v in val):
747
            self.desc.set_blocks_attr(name, [v.desc for v in val])
Q
Qiyang Min 已提交
748 749 750 751 752
        elif isinstance(val, core.BlockDesc) or \
                isinstance(val, core.ProgramDesc):
            self.desc.set_serialized_attr(name, val.serialize_to_string())
        else:
            self.desc.set_attr(name, val)
Y
yuyang18 已提交
753

F
fengjiayi 已提交
754 755 756 757 758
    @property
    def attr_names(self):
        return self.desc.attr_names()

    def attr(self, name):
759
        """
760 761
        Get the attribute by name.

762
        Args:
763
            name(str): the attribute name.
764

765 766
        Returns:
            bool|int|str|float|list: The attribute value. The return value
767 768
            can be any valid attribute type.
        """
F
fengjiayi 已提交
769
        return self.desc.attr(name)
Y
Yu Yang 已提交
770

F
fengjiayi 已提交
771
    def block_attr(self, name):
772
        """
773
        Get the block attribute by name.
774

775 776
        Args:
            name(str): the attribute name.
777

778 779
        Returns:
            int: the block index.
780
        """
F
fengjiayi 已提交
781
        return self.desc.block_attr(name)
Y
Yu Yang 已提交
782

J
JiayiFeng 已提交
783
    def all_attrs(self):
F
fengjiayi 已提交
784
        """
785 786 787 788
        Get the attribute dict.

        Returns:
            dict: The Operator's attribute dict.
F
fengjiayi 已提交
789 790 791 792 793 794 795 796 797 798
        """
        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
799

Y
Yu Yang 已提交
800
class Block(object):
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
    """
    In Fluid, a Program is consistence of multi-Block, and Block stores
    VarDesc and OpDesc. In a specific Block, a VarDesc have a unique name.
    One block could have some child blocks, and child block's name scopes
    should inherit the parent's so that OpDesc in child block can reference
    a VarDesc that is stored in the parent block.
    Please reference the framework.proto for details.

    Args:
        program(Program): The Program that the Block belongs to.
        idx(int): The block's id in the Program.

    Notes:
        The constructor of Block should not be invoked directly. Please
        use `Program.create_block()` to create a block.

    Examples:
        .. code-block:: python

            cur_program = Program()
            cur_block = cur_program.current_block()
            var = cur_block.create_var(name="X",
                                       shape=[-1, 23, 48],
                                       dtype='float32')
            cur_block.append_op(type="abs",
                                inputs={"X": [var]},
                                outputs={"Out": [var]})
    """

Y
Yu Yang 已提交
830
    def __init__(self, program, idx):
Y
Yu Yang 已提交
831
        self.desc = program.desc.block(idx)
832
        self.vars = collections.OrderedDict()  # var_name --> var
Q
qiaolongfei 已提交
833
        self.ops = list()  # operator list
Y
Yu Yang 已提交
834
        self.program = program
835
        self.removed_vars = collections.OrderedDict()
Y
Yu Yang 已提交
836

837
    def __str__(self):
Y
Yang Yang(Tony) 已提交
838 839
        return self.to_string(True)

F
fengjiayi 已提交
840 841
    def to_string(self, throw_on_error, with_details=False):
        """
842 843
        Get debug string.

F
fengjiayi 已提交
844 845
        Args:
            throw_on_error(bool): raise exception when self is not initialized
846
                when throw_on_error is True.
F
update  
fengjiayi 已提交
847
            with_details(bool): more details about variables and parameters
848 849
                (e.g. trainable, optimize_attr, ...) will be printed when
                with_details is True. Default False.
F
fengjiayi 已提交
850

851 852
        Returns:
            str: The debug string.
F
fengjiayi 已提交
853 854 855 856
        """
        assert isinstance(throw_on_error, bool) and isinstance(with_details,
                                                               bool)
        if with_details:
F
fengjiayi 已提交
857
            re_add_indent = re.compile(r"\n(.)")
F
fengjiayi 已提交
858 859
            res_str = "blocks {\n  idx: %d\n  parent_idx: %d" % (
                self.idx, self.parent_idx)
860
            for var in list(self.vars.values()):
F
fengjiayi 已提交
861
                res_str += "\n  vars {\n    %s  }" % re_add_indent.sub(
F
update  
fengjiayi 已提交
862
                    r"\n    \1", var.to_string(throw_on_error, with_details))
F
fengjiayi 已提交
863
            for op in self.ops:
F
fengjiayi 已提交
864 865
                res_str += "\n  ops {\n    %s  }" % re_add_indent.sub(
                    r"\n    \1", op.to_string(throw_on_error))
F
fengjiayi 已提交
866 867 868
            res_str += "\n}"
        else:
            protostr = self.desc.serialize_to_string()
869 870
            proto = framework_pb2.BlockDesc.FromString(
                six.binary_type(protostr))
F
fengjiayi 已提交
871 872
            res_str = _debug_string_(proto, throw_on_error)
        return res_str
873 874 875

    __repr__ = __str__

Y
Yu Yang 已提交
876 877
    @property
    def parent_idx(self):
Y
Yu Yang 已提交
878
        return self.desc.parent
Y
Yu Yang 已提交
879

Y
Yu Yang 已提交
880 881 882 883
    @property
    def forward_block_idx(self):
        return self.desc.get_forward_block_idx()

W
Wu Yi 已提交
884
    def _set_forward_block_idx(self, idx):
885 886 887 888 889 890 891 892 893
        """
        Set the forward block Idx.

        Args:
            idx(int): the block index.

        Returns:
            None
        """
W
Wu Yi 已提交
894
        self.desc._set_forward_block_idx(idx)
Y
Yu Yang 已提交
895

Y
Yu Yang 已提交
896 897
    @property
    def idx(self):
Y
Yu Yang 已提交
898
        return self.desc.id
Y
Yu Yang 已提交
899

Q
Qiao Longfei 已提交
900
    def var(self, name):
901 902 903 904 905 906 907 908 909 910 911 912 913
        """
        Get a Variable by name from this block.

        Args:
            name(str): the Variable's name.

        Raises:
            ValueError: The If input's type is not str, or this block
                doesn't have a Variable with the giving name.

        Returns:
            Variable: the Variable with the giving name.
        """
914 915 916 917 918
        if not issubclass(name.__class__, six.string_types):
            if not isinstance(name, six.binary_type):
                raise TypeError(
                    "var require string as parameter, but get %s instead." %
                    (type(name)))
Y
Yu Yang 已提交
919 920
        v = self.vars.get(name, None)
        if v is None:
Q
Qiao Longfei 已提交
921
            raise ValueError("var %s not in this block" % name)
Y
Yu Yang 已提交
922
        return v
Q
Qiao Longfei 已提交
923

W
Wu Yi 已提交
924
    def _var_recursive(self, name):
925 926 927 928 929 930 931 932 933 934 935 936 937
        """
        Get a Variable by name from this block recursively.

        Args:
            name(str): the Variable's name.

        Raises:
            ValueError: this block and this parent block doesn't
                have a Variable with the giving name.

        Returns:
            Variable: the Variable with the giving name.
        """
Y
Yu Yang 已提交
938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963
        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
964

Q
Qiao Longfei 已提交
965
    def all_parameters(self):
966
        return list(self._iter_parameters())
967

968
    def _iter_parameters(self):
969
        return (item[1] for item in list(self.vars.items())
970
                if isinstance(item[1], Parameter))
Q
Qiao Longfei 已提交
971

Y
Yu Yang 已提交
972
    def create_var(self, *args, **kwargs):
973
        var = Variable(block=self, *args, **kwargs)
974 975
        if 'initializer' in kwargs:
            kwargs['initializer'](var, self)
Q
Qiao Longfei 已提交
976
        return var
Y
Yu Yang 已提交
977

Q
Qiao Longfei 已提交
978 979 980
    def has_var(self, name):
        return name in self.vars

W
Wu Yi 已提交
981
    def _rename_var(self, name, new_name):
T
typhoonzero 已提交
982 983
        """
        Rename variable in vars and ops' inputs and outputs
984 985 986 987 988 989 990 991 992 993 994 995

        Args:
            name(str): the name that need to be renamed.
            new_name(str): the name that need to rename to.

        Raises:
            ValueError: If this block doesn't have this the giving name,
                or the type of the var with the giving name is not Parameter
                or Variable.

        Returns:
            Variable: the Variable with the giving name.
T
typhoonzero 已提交
996 997
        """
        if not self.has_var(name):
998
            raise ValueError("var %s is not in current block" % name)
T
wip  
typhoonzero 已提交
999 1000
        v = self.var(name)
        if type(v) == Parameter:
T
typhoonzero 已提交
1001
            var_type = "Parameter"
T
wip  
typhoonzero 已提交
1002 1003 1004 1005 1006 1007 1008
            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
1009
            var_type = "Variable"
T
wip  
typhoonzero 已提交
1010 1011 1012 1013
            error_clip = v.error_clip
            stop_gradient = v.stop_gradient
        else:
            raise ValueError("unsupported var type: %s", type(v))
T
typhoonzero 已提交
1014
        orig_var_type = v.type
W
Wu Yi 已提交
1015 1016
        self.desc._rename_var(name, new_name)
        # NOTE: v is destroyed by C++ after calling _rename_var.
T
wip  
typhoonzero 已提交
1017
        d = self.desc.find_var(new_name)
T
typhoonzero 已提交
1018
        if var_type == "Parameter":
T
wip  
typhoonzero 已提交
1019 1020 1021 1022
            var = Parameter(
                self,
                d.shape(),
                d.dtype(),
T
typhoonzero 已提交
1023
                type=orig_var_type,
T
wip  
typhoonzero 已提交
1024 1025 1026 1027 1028 1029 1030
                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
1031
        elif var_type == "Variable":
T
wip  
typhoonzero 已提交
1032 1033
            var = Variable(
                self,
T
typhoonzero 已提交
1034
                type=orig_var_type,
T
wip  
typhoonzero 已提交
1035 1036 1037 1038
                name=new_name,
                error_clip=error_clip,
                stop_gradient=stop_gradient)

W
Wu Yi 已提交
1039
        # rename the python side, _sync_with_cpp will only add
T
wip  
typhoonzero 已提交
1040 1041 1042
        # new vars/ops to python side.
        self.vars[new_name] = var
        del self.vars[name]
W
Wu Yi 已提交
1043
        self._sync_with_cpp()
1044
        return var
T
typhoonzero 已提交
1045

W
Wu Yi 已提交
1046 1047 1048
    def _remove_var(self, name):
        self._sync_with_cpp()
        self.desc._remove_var(name)
1049 1050
        del self.vars[name]

Y
Yu Yang 已提交
1051 1052
    def create_parameter(self, *args, **kwargs):
        global_block = self.program.global_block()
Q
Qiao Longfei 已提交
1053
        param = Parameter(global_block, *args, **kwargs)
1054 1055
        if 'initializer' in kwargs:
            kwargs['initializer'](param, self)
Q
Qiao Longfei 已提交
1056
        return param
Y
Yu Yang 已提交
1057

Y
Yu Yang 已提交
1058
    def append_op(self, *args, **kwargs):
1059 1060 1061 1062 1063 1064
        """
        Appends a new Operator according to the giving arguments.

        Returns:
            Operator: the append Operator.
        """
Y
Yu Yang 已提交
1065
        op_desc = self.desc.append_op()
1066
        op = Operator(block=self, desc=op_desc, *args, **kwargs)
Y
Yu Yang 已提交
1067 1068 1069
        self.ops.append(op)
        return op

W
Wu Yi 已提交
1070
    def _insert_op(self, index, *args, **kwargs):
1071 1072 1073 1074 1075 1076 1077 1078 1079
        """
        Insert a Operator according to the giving arguments.

        Args:
            index(int): the place that the operator to insert.

        Returns:
            Operator: the insert Operator.
        """
W
Wu Yi 已提交
1080 1081
        self._sync_with_cpp()
        op_desc = self.desc._insert_op(index)
Q
qiaolongfei 已提交
1082 1083 1084 1085
        op = Operator(block=self, desc=op_desc, *args, **kwargs)
        self.ops.insert(index, op)
        return op

W
Wu Yi 已提交
1086
    def _remove_op(self, index):
1087 1088 1089 1090 1091 1092 1093 1094 1095
        """
        Remove the specific position operator.

        Args:
            index(int): the position that the operator to insert.

        Returns:
            None
        """
W
Wu Yi 已提交
1096 1097
        self._sync_with_cpp()
        self.desc._remove_op(index, index + 1)
1098 1099
        del self.ops[index]

W
Wu Yi 已提交
1100
    def _slice_ops(self, start, end):
1101 1102 1103 1104 1105 1106 1107 1108 1109 1110
        """
        Return the Operator between start and end.

        Args:
            start(int): the start position.
            end(int): the end position.

        Returns:
            list: the Operators between start and end.
        """
Q
qiaolongfei 已提交
1111
        return self.ops[start:end]
Y
Yancey1989 已提交
1112

W
Wu Yi 已提交
1113 1114
    def _prepend_op(self, *args, **kwargs):
        op_desc = self.desc._prepend_op()
Y
Yu Yang 已提交
1115
        op = Operator(self, op_desc, *args, **kwargs)
Q
qiaolongfei 已提交
1116
        self.ops.insert(0, op)
Y
Yu Yang 已提交
1117 1118
        return op

W
Wu Yi 已提交
1119
    def _sync_with_cpp(self):
1120
        """
1121 1122
        Sync from the desc on the c++ end. This method is used to synchronize
        the c++ desc instance generated by backward.
1123
        """
Q
Qiao Longfei 已提交
1124 1125 1126 1127 1128
        # 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())

1129
        # sync variables removed from c++ end
1130
        for var in list(self.vars.keys()):
1131 1132 1133
            if not self.desc.find_var(var):
                self.vars.pop(var)

Q
Qiao Longfei 已提交
1134
        # sync operators from cpp
1135 1136 1137 1138
        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
1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154
        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
1155 1156 1157 1158 1159

        # 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
1160
            self.ops.insert(0, op)
Q
Qiao Longfei 已提交
1161 1162 1163 1164 1165 1166 1167

        # 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)

1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180
        # 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
1181 1182 1183 1184
        assert len(self.ops) == len(ops_in_cpp)
        for index in range(len(self.ops)):
            assert self.ops[index].desc == ops_in_cpp[index]

W
Wu Yi 已提交
1185
    def _copy_param_info_from(self, other):
1186
        """
1187 1188
        Copy the information of parameters from the other block.

1189
        Args:
1190 1191 1192 1193 1194
            other(Block): the other block.

        Raises:
            ValueError: If type of input is not Block, or the `other` and this
                block is not in the same topology.
1195 1196 1197 1198 1199

        Returns:
            None
        """
        if not isinstance(other, Block):
W
Wu Yi 已提交
1200 1201
            raise TypeError(
                "_copy_param_info_from should be invoked with Block")
1202
        for p in other._iter_parameters():
1203 1204 1205
            assert isinstance(p, Parameter)
            v = self.vars.get(p.name, None)
            if v is None:
W
Wu Yi 已提交
1206
                raise ValueError("_copy_param_info_from should be invoked with "
1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218
                                 "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
1219
                gradient_clip_attr=p.gradient_clip_attr,
F
fengjiayi 已提交
1220
                error_clip=p.error_clip,
1221 1222 1223
                name=v.name)
            self.vars[new_p.name] = new_p

W
Wu Yi 已提交
1224
    def _clone_variable(self, var):
1225 1226
        """
        Clone a variable into current block.
1227

1228 1229 1230 1231
        Args:
            var: the variable to be cloned.

        Returns:
1232
            Variable: the new  variable cloned from 'var' in current block.
1233 1234
        """
        assert isinstance(var, Variable)
T
update  
typhoonzero 已提交
1235 1236 1237 1238 1239
        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
tangwei12 已提交
1240 1241
        elif var.type == core.VarDesc.VarType.RAW:
            ret_var = self.create_var(
T
tangwei12 已提交
1242
                name=var.name, persistable=var.persistable, type=var.type)
T
typhoonzero 已提交
1243 1244 1245 1246 1247 1248
        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,
F
fengjiayi 已提交
1249 1250
                persistable=True,
                is_data=var.is_data)
T
update  
typhoonzero 已提交
1251 1252 1253 1254 1255 1256 1257
        else:
            ret_var = self.create_var(
                name=var.name,
                shape=var.shape,
                dtype=var.dtype,
                type=var.type,
                lod_level=var.lod_level,
F
fengjiayi 已提交
1258 1259
                persistable=True,
                is_data=var.is_data)
T
update  
typhoonzero 已提交
1260
        return ret_var
1261

Y
Yu Yang 已提交
1262 1263

class Program(object):
D
dzhwinter 已提交
1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274
    """
    Python Program. Beneath it is a ProgramDesc, which is used for
    create c++ Program. A program is a self-contained programing
    language like container. It has at least one Block, when the
    control flow op like conditional_block, while_op is included,
    it will contains nested block.
    Please reference the framework.proto for details.

    Notes: we have default_startup_program and default_main_program
    by default, a pair of them will shared the parameters.
    The default_startup_program only run once to initialize parameters,
Y
yuyang18 已提交
1275
    default_main_program run in every mini batch and adjust the weights.
D
dzhwinter 已提交
1276 1277

    Returns:
Y
yuyang18 已提交
1278
        A empty program.
D
dzhwinter 已提交
1279 1280

    Examples:
Y
yuyang18 已提交
1281 1282 1283 1284 1285 1286
        >>> main_program = fluid.Program()
        >>> startup_program = fluid.Program()
        >>> with fluid.program_guard(main_program=main_program, startup_program=startup_program):
        >>>     fluid.layers.data(name="x", shape=[-1, 784], dtype='float32')
        >>>     fluid.layers.data(name="y", shape=[-1, 1], dtype='int32')
        >>>     fluid.layers.fc(name="fc", shape=[10], dtype='float32', act="relu")
D
dzhwinter 已提交
1287 1288 1289

    """

1290 1291
    def __init__(self):
        self.desc = core.ProgramDesc()
Y
Yu Yang 已提交
1292 1293
        self.blocks = [Block(self, 0)]
        self.current_block_idx = 0
D
dzhwinter 已提交
1294
        self._seed = 0
Y
yuyang18 已提交
1295
        self._current_role = core.op_proto_and_checker_maker.OpRole.Forward
Y
yuyang18 已提交
1296
        self._op_role_var = []
Y
yuyang18 已提交
1297 1298 1299

    @property
    def op_role(self):
Y
yuyang18 已提交
1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312
        """
        The operator role. In a enum {Forward, Backward, Optimize}.

        Notes: this is a low level API. It is used only for ParallelExecutor to
        duplicate or schedule operator to devices.

        For example, the forward operator should be executed on every device.
        The backward operator should be executed on every device and the
        parameter gradient of backward (use :code:`op_role_var` to get this
        variable) operator should be merged to one device. The optimization
        operators should be executed on only one device and broadcast the
        optimization result, i.e., the new parameter, to every other device.
        """
Y
yuyang18 已提交
1313 1314 1315 1316 1317 1318 1319 1320
        return self._current_role

    @op_role.setter
    def set_op_role(self, role):
        self._current_role = role

    @property
    def op_role_var(self):
Y
yuyang18 已提交
1321 1322 1323 1324 1325 1326 1327
        """
        The auxiliary variables for :code:`op_role` property.

        See Also: :code:`Program.op_role`'s documentation for details.

        Notes: This is a very low-level API. Users should not use it directly.
        """
Y
yuyang18 已提交
1328 1329 1330 1331
        return self._op_role_var

    @op_role_var.setter
    def set_op_role_var(self, var_name):
Y
yuyang18 已提交
1332
        self._op_role_var = [var_name]
Y
yuyang18 已提交
1333 1334

    @contextlib.contextmanager
1335
    def optimized_guard(self, param_and_grads):
Y
yuyang18 已提交
1336 1337 1338 1339 1340 1341 1342
        """
        A with guard to set :code:`Optimization` :code:`OpRole` and
        :code:`OpRoleVar` automatically.

        Notes: This is a very low level API. Users should not use it directly.

        Args:
1343
            param_and_grads(list): The variables (names) to be optimized.
Y
yuyang18 已提交
1344 1345 1346 1347

        Examples:

            >>> p, g = backward(...)
1348
            >>> with program.optimized_guard([p,g]):
Y
yuyang18 已提交
1349 1350
            >>>     p = p - 0.001 * g
        """
Y
yuyang18 已提交
1351 1352
        OpRole = core.op_proto_and_checker_maker.OpRole
        self._current_role = OpRole.Optimize
1353 1354 1355 1356
        self._op_role_var = [
            var.name if isinstance(var, Variable) else var
            for var in param_and_grads
        ]
Y
yuyang18 已提交
1357
        yield
Y
yuyang18 已提交
1358
        self._op_role_var = []
Y
yuyang18 已提交
1359
        self._current_role = OpRole.Forward
Y
Yu Yang 已提交
1360

1361
    def __str__(self):
Y
yuyang18 已提交
1362 1363 1364 1365 1366 1367 1368 1369 1370
        """
        Get the protobuf debug string of this Program.

        Returns:
            (str): The protobuf debug string.

        Raises:
            ValueError: If any of required fields is not set.
        """
Y
Yang Yang(Tony) 已提交
1371 1372
        return self.to_string(True)

F
fengjiayi 已提交
1373 1374 1375
    def to_string(self, throw_on_error, with_details=False):
        """
        To debug string.
Y
yuyang18 已提交
1376

F
fengjiayi 已提交
1377
        Args:
Y
yuyang18 已提交
1378 1379
            throw_on_error(bool): raise Value error when any of required fields
                is not set.
F
fengjiayi 已提交
1380

Y
yuyang18 已提交
1381 1382 1383 1384 1385 1386 1387 1388 1389 1390
            with_details(bool): True if more details about variables and
                parameters, e.g., :code:`trainable`, :code:`optimize_attr`, need
                to print.

        Returns
            (str): The debug string.

        Raises:
            ValueError: If any of required fields is not set and throw_on_error is
                True.
F
fengjiayi 已提交
1391 1392 1393 1394 1395 1396 1397 1398 1399 1400

        """
        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()
1401 1402
            proto = framework_pb2.ProgramDesc.FromString(
                six.binary_type(protostr))
F
fengjiayi 已提交
1403 1404
            res_str = _debug_string_(proto, throw_on_error)
        return res_str
1405

1406
    def get_desc(self):
Y
yuyang18 已提交
1407 1408 1409 1410 1411 1412 1413
        """
        Get the C++ side of `ProgramDesc` object pointer. The C++ object is
        exposed by :code:`pybind`.

        Notes: This is a very low level API. Users should not use this API
        directly.
        """
1414 1415
        return self.desc

1416
    def clone(self, for_test=False):
Y
yuyang18 已提交
1417 1418 1419
        """
        Create a new, duplicated program.

1420

Y
yuyang18 已提交
1421 1422 1423 1424
        Some operators, e.g., :code:`batch_norm`, behave differently between
        training and testing. They have an attribute, :code:`is_test`, to
        control this behaviour. This method will change the :code:`is_test`
        attribute of them to :code:`True` when :code:`for_test=True`.
1425

Y
yuyang18 已提交
1426 1427 1428 1429
        * Set for_test to False when we want to clone the program for training.
        * Set for_test to True when we want to clone the program for testing.

        Notes: This API DOES NOT prune any operator. Use
L
Luo Tao 已提交
1430 1431 1432 1433 1434
        :code:`clone(for_test=True)` before backward and optimization please. e.g.

            >>> test_program = fluid.default_main_program().clone(for_test=True)
            >>> optimizer = fluid.optimizer.Momentum(learning_rate=0.01, momentum=0.9)
            >>> optimizer.minimize()
1435 1436

        Args:
Y
yuyang18 已提交
1437 1438
            for_test(bool): True if change the :code:`is_test` attribute of
                operators to :code:`True`.
1439

D
dzhwinter 已提交
1440
        Returns:
Y
yuyang18 已提交
1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493
            Program: The new, duplicated Program object.

        Examples:

            1. To clone a test program, the sample code is:

            >>> import paddle.fluid as fluid
            >>> train_program = fluid.Program()
            >>> startup_program = fluid.Program()
            >>> with fluid.program_guard(train_program, startup_program):
            >>>     img = fluid.layers.data(name='image', shape=[784])
            >>>     hidden = fluid.layers.fc(input=img, size=200, act='relu')
            >>>     hidden = fluid.layers.dropout(hidden, dropout_prob=0.5)
            >>>     loss = fluid.layers.cross_entropy(
            >>>                 input=fluid.layers.fc(hidden, size=10, act='softmax'),
            >>>                 label=fluid.layers.data(name='label', shape=[1], dtype='int64'))
            >>>
            >>> test_program = train_program.clone(for_test=True)
            >>>
            >>> sgd = fluid.optimizer.SGD(learning_rate=1e-3)
            >>> with fluid.program_guard(train_program, startup_program):
            >>>     sgd.minimize(loss)

            2. The :code:`clone` method can be avoid if you create program for
            training and program for testing individually.

            >>> import paddle.fluid as fluid
            >>>
            >>> def network(is_test):
            >>>     img = fluid.layers.data(name='image', shape=[784])
            >>>     hidden = fluid.layers.fc(input=img, size=200, act='relu')
            >>>     hidden = fluid.layers.dropout(hidden, dropout_prob=0.5, is_test=is_test)
            >>>     loss = fluid.layers.cross_entropy(
            >>>                 input=fluid.layers.fc(hidden, size=10, act='softmax'),
            >>>                 label=fluid.layers.data(name='label', shape=[1], dtype='int64'))
            >>>     return loss
            >>>
            >>> train_program = fluid.Program()
            >>> startup_program = fluid.Program()
            >>> test_program = fluid.Program()
            >>>
            >>> with fluid.program_guard(train_program, startup_program):
            >>>     with fluid.unique_name.guard():
            >>>         loss = network(is_test=False)
            >>>         sgd = fluid.optimizer.SGD(learning_rate=1e-3)
            >>>         sgd.minimize(loss)
            >>>
            >>> # the test startup program is not used.
            >>> with fluid.program_guard(test_program, fluid.Program()):
            >>>     with fluid.unique_name.guard():
            >>>         loss = network(is_test=True)

            The two code snippets above will generate same programs.
1494 1495
        """
        if for_test:
1496
            p = self.inference_optimize()
1497
        else:
1498
            p = Program()
1499
            p.desc = core.ProgramDesc(self.desc)
1500
            p.blocks = [Block(p, i) for i in range(self.desc.num_blocks())]
W
Wu Yi 已提交
1501
            p._sync_with_cpp()
1502

W
Wu Yi 已提交
1503
        p._copy_param_info_from(self)
F
fengjiayi 已提交
1504
        p.copy_data_info_from(self)
Y
Yu Yang 已提交
1505
        return p
1506

1507
    def prune(self, targets):
Y
yuyang18 已提交
1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522
        """
        Prune operators and variables which are not needed to generate
        :code:`targets`.

        Notes: This is a very low level API. Users should not use this API
        directly. This API is in flux and not stable.

        Args:
            targets(list|Variable|Operator): A list of variables or operators
                need to be pruned

        Returns:
            Program:  A new, pruned program.

        """
1523 1524 1525 1526 1527 1528
        if not isinstance(targets, list):
            targets = [targets]
        targets_idx = []
        for t in targets:
            if not isinstance(t, Operator):
                if isinstance(t, Variable):
1529 1530
                    # After transpiler processing, the op that output this
                    # variable maybe has been changed, so t.op is not reliable
1531
                    # and we need to find the current op that generate this
1532 1533 1534 1535 1536 1537 1538 1539
                    # variable here.
                    t.op = None
                    global_block = self.global_block()
                    for idx, op in enumerate(global_block.ops):
                        if t.name in op.output_arg_names:
                            t.op = op
                            break

1540
                    t = t.op
1541 1542 1543 1544
                    if t is None:
                        raise ValueError(
                            "The target variable must have an "
                            "associated operator that generates it.")
1545
                else:
1546 1547
                    raise ValueError("All targets of prune() can only be "
                                     "Variable or Operator.")
1548 1549 1550 1551

            targets_idx.append([t.block.idx, t.idx])
        res = Program()
        res.desc = core.prune(self.desc, targets_idx)
1552
        res.blocks = [Block(res, i) for i in range(res.desc.num_blocks())]
W
Wu Yi 已提交
1553
        res._sync_with_cpp()
1554 1555
        return res

1556
    def inference_optimize(self):
Y
yuyang18 已提交
1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567
        """
        This method will create a new program and change the :code:`is_test`
        attribute of operators to :code:`True`. All the :code:`Parameter`
        information will be lost.

        Notes: This API is a very low level API. Use
        :code:`Program.clone(for_test=True)` instead.

        Returns:
            Program: The new program.
        """
1568 1569
        # this is an alternative implement before
        # core.inference_optimize being fixed.
1570
        res = Program()
1571
        res.desc = core.ProgramDesc(self.desc)
1572
        for i in range(res.desc.num_blocks()):
1573
            block = res.desc.block(i)
1574
            for j in range(block.op_size()):
1575 1576 1577
                op = block.op(j)
                if op.has_attr('is_test'):
                    op.set_attr('is_test', True)
1578
        res.blocks = [Block(res, i) for i in range(res.desc.num_blocks())]
W
Wu Yi 已提交
1579
        res._sync_with_cpp()
1580 1581
        return res

1582 1583
    @staticmethod
    def parse_from_string(binary_str):
Y
yuyang18 已提交
1584 1585 1586 1587 1588 1589 1590
        """
        Deserialize a program desc from protobuf binary string.

        Notes: All information about parameters will be lost after serialization
        and deserialization.

        Args:
1591
            binary_str_type(str): The binary prootbuf string.
Y
yuyang18 已提交
1592 1593 1594 1595

        Returns:
            Program: A deserialized program desc.
        """
1596 1597
        p = Program()
        p.desc = core.ProgramDesc(binary_str)
1598
        p.blocks = [Block(p, i) for i in range(p.desc.num_blocks())]
W
Wu Yi 已提交
1599
        p._sync_with_cpp()
1600
        return p
Y
Yu Yang 已提交
1601

D
dzhwinter 已提交
1602 1603
    @property
    def random_seed(self):
Y
yuyang18 已提交
1604 1605 1606 1607 1608 1609
        """
        The default random seed for random operators in Program. Zero means get
        the random seed from random device.

        Notes: It must be set before the operators have been added.
        """
D
dzhwinter 已提交
1610 1611
        return self._seed

Q
qiaolongfei 已提交
1612 1613
    @property
    def num_blocks(self):
Y
yuyang18 已提交
1614 1615 1616
        """
        The number of blocks in this program.
        """
Q
qiaolongfei 已提交
1617 1618
        return self.desc.num_blocks()

D
dzhwinter 已提交
1619 1620 1621 1622 1623 1624
    @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
1625
    def __repr__(self):
1626
        return self.__str__()
1627

Y
Yu Yang 已提交
1628
    def global_block(self):
Y
yuyang18 已提交
1629 1630 1631
        """
        Get the first block of this program.
        """
Y
Yu Yang 已提交
1632 1633
        return self.blocks[0]

Q
Qiao Longfei 已提交
1634
    def block(self, index):
Y
yuyang18 已提交
1635 1636 1637 1638 1639 1640 1641 1642
        """
        Get the :code:`index` block of this program
        Args:
            index(int): The index of block to get

        Returns:
            Block: The :code:`index` block
        """
Q
Qiao Longfei 已提交
1643 1644
        return self.blocks[index]

Y
Yu Yang 已提交
1645
    def current_block(self):
Y
yuyang18 已提交
1646 1647 1648 1649
        """
        Get the current block. The :code:`current` block is the block to append
        operators.
        """
Y
Yu Yang 已提交
1650 1651
        return self.blocks[self.current_block_idx]

F
update  
fengjiayi 已提交
1652
    def create_block(self, parent_idx=None):
Y
yuyang18 已提交
1653 1654 1655 1656 1657 1658 1659 1660 1661 1662
        """
        Create a new block with the :code:`parent_idx` and change the current block
        to new block.

        Args:
            parent_idx(int): The parent block index.

        Returns:
            Block: The new block.
        """
Y
Yu Yang 已提交
1663
        new_block_idx = len(self.blocks)
F
update  
fengjiayi 已提交
1664 1665 1666
        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
1667 1668 1669 1670 1671
        self.current_block_idx = new_block_idx
        self.blocks.append(Block(self, self.current_block_idx))
        return self.current_block()

    def rollback(self):
Y
yuyang18 已提交
1672 1673 1674 1675 1676
        """
        Exit a code block, i.e., roll back to the parent block.
        Returns:
            None
        """
Y
Yu Yang 已提交
1677 1678
        self.current_block_idx = self.current_block().parent_idx

W
Wu Yi 已提交
1679
    def _sync_with_cpp(self):
Y
yuyang18 已提交
1680 1681 1682 1683 1684 1685 1686 1687 1688 1689
        """
        Synchronize Python instance to its binding C++ object instance.
        If the program is modified in C++ space, this method should be invoked.

        Notes: This is a very low level API. Users should not invoke it
        directly.

        Returns:
            None
        """
Q
Qiao Longfei 已提交
1690 1691 1692
        for block_idx in range(len(self.blocks), self.desc.num_blocks()):
            self.blocks.append(Block(self, block_idx))
        for block in self.blocks:
W
Wu Yi 已提交
1693
            block._sync_with_cpp()
Q
Qiao Longfei 已提交
1694

W
Wu Yi 已提交
1695
    def _copy_param_info_from(self, other):
1696
        """
1697
        Copy the information of parameters from other program.
D
dzhwinter 已提交
1698

Y
yuyang18 已提交
1699 1700 1701
        Notes: This is a very low level API. Users should not invoke it
        directly.

1702 1703 1704 1705 1706 1707 1708
        Args:
            other(Program): Other program

        Returns:
            None
        """
        if not isinstance(other, Program):
W
Wu Yi 已提交
1709
            raise TypeError("_copy_param_info_from should be invoked with "
1710 1711 1712
                            "Program")

        if len(self.blocks) != len(other.blocks):
W
Wu Yi 已提交
1713
            raise ValueError("_copy_param_info_from should be invoked with two "
1714
                             "program, with represent the same topology")
W
Wu Yi 已提交
1715
        self.global_block()._copy_param_info_from(other.global_block())
1716

F
fengjiayi 已提交
1717 1718 1719
    def copy_data_info_from(self, other):
        """
        Copy the information of data variables from other program.
D
dzhwinter 已提交
1720

Y
yuyang18 已提交
1721 1722 1723
        Notes: This is a very low level API. Users should not invoke it
        directly.

F
fengjiayi 已提交
1724 1725 1726 1727 1728 1729 1730
        Args:
            other(Program): Other program

        Returns:
            None
        """
        if not isinstance(other, Program):
W
Wu Yi 已提交
1731
            raise TypeError("_copy_param_info_from should be invoked with "
F
fengjiayi 已提交
1732 1733 1734
                            "Program")

        if len(self.blocks) != len(other.blocks):
W
Wu Yi 已提交
1735
            raise ValueError("_copy_param_info_from should be invoked with two "
F
fengjiayi 已提交
1736
                             "program, with represent the same topology")
1737
        for var in list(other.global_block().vars.values()):
F
fengjiayi 已提交
1738 1739 1740
            if var.is_data:
                self.global_block().var(var.name).is_data = True

1741
    def list_vars(self):
Y
yuyang18 已提交
1742 1743 1744 1745 1746 1747
        """
        Get all variables from this Program. A iterable object is returned.

        Returns:
            iterable: The generator will yield every variable in this program.
        """
1748
        for each_block in self.blocks:
1749
            for each_var in list(each_block.vars.values()):
1750 1751
                yield each_var

Y
Yu Yang 已提交
1752

Y
Yu Yang 已提交
1753
class Parameter(Variable):
1754
    """
1755
    Parameter is derived from Variable. A parameter is a persistable
1756
    Variable, and will be updated by optimizers after each iteration.
1757
    The training of a neural network is essentially the updating of
1758 1759
    its parameters.

1760
    Relative to a general Variable, a Parameter has several its own
1761 1762
    member variables:

1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774
    Args:
        trainable(bool): True if the parameter need to be updated after
            iterations.
        optimize_attr(map): Parameter attributes related with optimizing.
            Currently, it only contains 'learning_rate'.
            Default: {'learning_rate': 1.0}
        regularizer(WeightDecayRegularizer): The Regularizer which will
            be applied on the parameter. Default: None
        gradient_clip_attr(BaseGradientClipAttr): The gradint clip strategy
            which will be applied on the parameter. Default: None
        do_model_average(bool): True if the model average strategy will
            be applied on this parameter.
1775 1776
    """

Y
Yu Yang 已提交
1777 1778 1779 1780 1781 1782 1783 1784 1785 1786
    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")
1787 1788 1789

        Variable.__init__(
            self, block, persistable=True, shape=shape, dtype=dtype, **kwargs)
Y
Yu Yang 已提交
1790 1791 1792 1793
        self.trainable = kwargs.get('trainable', True)

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

1794 1795
        self.regularizer = kwargs.get('regularizer', None)

F
fengjiayi 已提交
1796
        self.gradient_clip_attr = kwargs.get('gradient_clip_attr', None)
Y
Yu Yang 已提交
1797

W
wanghaoshuang 已提交
1798
        self.do_model_average = kwargs.get('do_model_average', None)
W
wanghaoshuang 已提交
1799

F
fengjiayi 已提交
1800 1801 1802
    def __str__(self):
        return self.to_string(True)

F
update  
fengjiayi 已提交
1803 1804 1805
    def to_string(self, throw_on_error, with_details=False):
        """
        To debug string.
D
dzhwinter 已提交
1806

F
update  
fengjiayi 已提交
1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820
        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
1821
                               "gradient_clip_attr", "do_model_average")
F
update  
fengjiayi 已提交
1822
            for attr_name in additional_attr:
1823 1824
                res_str += "%s: %s\n" % (
                    attr_name, six.binary_type(getattr(self, attr_name)))
F
update  
fengjiayi 已提交
1825 1826
        else:
            res_str = Variable.to_string(self, throw_on_error, False)
F
fengjiayi 已提交
1827 1828 1829 1830
        return res_str

    __repr__ = __str__

Y
Yu Yang 已提交
1831

Y
Yu Yang 已提交
1832
# program is a global instance.
Y
Yu Yang 已提交
1833 1834
_main_program_ = Program()
_startup_program_ = Program()
1835

1836

1837
def default_startup_program():
Y
Yu Yang 已提交
1838
    """
Y
yuyang18 已提交
1839 1840 1841 1842 1843 1844 1845 1846 1847
    Get default/global startup program.

    The layer function in :code:`fluid.layers` will create parameters, readers,
    NCCL handles as global variables. The :code:`startup_program` will
    initialize them by the operators in startup program. The layer function will
    append these initialization operators into startup program.

    This method will return the :code:`default` or the :code:`current` startup
    program. Users can use :code:`fluid.program_guard` to switch program.
1848

Y
Yu Yang 已提交
1849 1850 1851
    Returns:
        Program: startup program
    """
Y
Yu Yang 已提交
1852
    return _startup_program_
1853

1854

1855
def default_main_program():
Y
Yu Yang 已提交
1856
    """
Y
yuyang18 已提交
1857 1858 1859 1860 1861 1862 1863 1864 1865
    Get default/global main program. The main program is used for training or
    testing.

    All layer function in :code:`fluid.layers` will append operators and
    variables to the :code:`default_main_program`.

    The :code:`default_main_program` is the default program in a lot of APIs.
    For example, the :code:`Executor.run()` will execute the
    :code:`default_main_program` when the program is not specified.
1866

Y
Yu Yang 已提交
1867 1868 1869
    Returns:
        Program: main program
    """
Y
Yu Yang 已提交
1870
    return _main_program_
Y
Yu Yang 已提交
1871 1872 1873 1874 1875


def switch_main_program(program):
    """
    Switch the main program to a new program.
1876

Y
Yu Yang 已提交
1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890
    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):
    """
1891
    Switch the startup program to a new program
Y
Yu Yang 已提交
1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906
    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):
    """
Y
yuyang18 已提交
1907 1908 1909
    Change the global main program and startup program with `with` statement.
    Layer functions in the Python `with` block will append operators and
    variables to the new main programs.
1910

Y
Yu Yang 已提交
1911
    Examples:
Y
yuyang18 已提交
1912 1913 1914 1915 1916 1917 1918 1919 1920 1921

        >>> import paddle.fluid as fluid
        >>> main_program = fluid.Program()
        >>> startup_program = fluid.Program()
        >>> with fluid.program_guard(main_program, startup_program):
        >>>     data = fluid.layers.data(...)
        >>>     hidden = fluid.layers.fc(...)

    Notes: The temporary :code:`Program` can be used if the user does not need
    to construct either of startup program or main program.
1922

Y
Yu Yang 已提交
1923
    Examples:
Y
yuyang18 已提交
1924 1925 1926 1927 1928 1929

        >>> import paddle.fluid as fluid
        >>> main_program = fluid.Program()
        >>> # does not care about startup program. Just pass a temporary value.
        >>> with fluid.program_guard(main_program, fluid.Program()):
        >>>     data = ...
1930

Y
Yu Yang 已提交
1931
    Args:
Y
yuyang18 已提交
1932
        main_program(Program): New main program inside `with` statement.
1933
        startup_program(Program): New startup program inside `with` statement.
Y
Yu Yang 已提交
1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946
            None means do not change startup program.
    """
    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
1947 1948 1949 1950


def get_var(name, program=None):
    """
Y
yuyang18 已提交
1951
    Get a variable by name from the global block of a program.
F
fengjiayi 已提交
1952

X
xuwei06 已提交
1953 1954 1955
    Args:
        name(str): name of the variable
        program(Program|None): program object.
T
tangwei12 已提交
1956
        If None, default_global_program() will be used.
X
xuwei06 已提交
1957 1958 1959 1960 1961 1962 1963

    Returns:
        Variable
    """
    if program is None:
        program = default_main_program()
    assert isinstance(name, str)
1964
    assert isinstance(program, Program)
X
xuwei06 已提交
1965 1966

    return program.global_block().var(name)