framework.py 65.7 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

M
minqiyang 已提交
22
from . import compat as cpt
23
from .proto import framework_pb2
24 25
try:
    from . import core
26
except ImportError as e:
27 28 29 30
    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
31
    directory. The original error is: \n""" + e.message)
32
except Exception as e:
33
    raise e
34
from . import unique_name
Y
Yu Yang 已提交
35

36
__all__ = [
37 38
    'Program',
    'Operator',
F
fengjiayi 已提交
39
    'Parameter',
40 41 42
    'default_startup_program',
    'default_main_program',
    'program_guard',
X
xuwei06 已提交
43
    'get_var',
44
]
Y
Yu Yang 已提交
45

Q
qiaolongfei 已提交
46 47 48 49 50 51 52 53
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):
    """
54 55
    Returns:
        str: gradient name for a certain var name
Q
qiaolongfei 已提交
56 57 58
    """
    return var_name + GRAD_VAR_SUFFIX

Y
Yu Yang 已提交
59

60
def convert_np_dtype_to_dtype_(np_dtype):
61 62
    """
    Convert the data type in numpy to the data type in Paddle
63

64
    Args:
65
        np_dtype(np.dtype): the data type in numpy.
66

67 68
    Returns:
        core.VarDesc.VarType: the data type in Paddle.
69 70

    """
71 72
    dtype = np.dtype(np_dtype)
    if dtype == np.float32:
73
        return core.VarDesc.VarType.FP32
74
    elif dtype == np.float64:
75
        return core.VarDesc.VarType.FP64
76
    elif dtype == np.float16:
77
        return core.VarDesc.VarType.FP16
78
    elif dtype == np.int32:
79
        return core.VarDesc.VarType.INT32
80
    elif dtype == np.int16:
81
        return core.VarDesc.VarType.INT16
82
    elif dtype == np.int64:
83
        return core.VarDesc.VarType.INT64
84
    elif dtype == np.bool:
85
        return core.VarDesc.VarType.BOOL
86 87
    elif dtype == np.uint16:
        return core.VarDesc.VarType.INT16
88 89
    elif dtype == np.uint8:
        return core.VarDesc.VarType.UINT8
90
    else:
M
minqiyang 已提交
91
        raise ValueError("Not supported numpy dtype %s" % dtype)
92 93 94


def dtype_is_floating(dtype):
95 96 97
    """
    Check the data type is floating or not.
    Args:
98
        dtype(np.dtype|core.VarDesc.VarType): data type.
99 100 101 102 103
            Could be numpy format or Paddle format

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

    """
104
    if not isinstance(dtype, core.VarDesc.VarType):
105 106
        dtype = convert_np_dtype_to_dtype_(dtype)

107 108 109 110
    return dtype in [
        core.VarDesc.VarType.FP16, core.VarDesc.VarType.FP32,
        core.VarDesc.VarType.FP64
    ]
111 112


Y
Yang Yang(Tony) 已提交
113
def _debug_string_(proto, throw_on_error=True):
114 115 116 117 118 119 120 121 122 123 124
    """
    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
125
    error_fields = list()
Y
Yang Yang(Tony) 已提交
126
    if not proto.IsInitialized(error_fields) and throw_on_error:
C
caoying03 已提交
127 128
        raise ValueError("{0} are not initialized.\nThe message is {1}:\n".
                         format(error_fields, proto))
Y
Yu Yang 已提交
129 130 131
    return proto.__str__()


Y
Yu Yang 已提交
132
class Variable(object):
133
    """
134 135 136
    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
137
    two variables in different blocks could have the same name.
138

139 140
    There are many kinds of variables. Each kind of them has its own attributes
    and usages. Please reference the framework.proto for details.
141

142
    Most of a Variable's member variables can be setted to be None. It mean
143
    it is not available or will be specified later.
144 145

    Args:
146
        block(Block): The block that the variable belongs to.
147 148
        type(core.VarDesc.VarType): Variable type. Please reference the
            framework.proto for details.
149 150
        name(str|None): The name of the variable. If setted None, it will be
            generated automatically. Default: None
151
        shape(tuple|list|None): The shape of the variable. -1 means the batch size.
152
            Some kinds of variable do not contain shape, just set it to None.
153 154 155
            Default: None
        dtype(np.dtype|core.VarDesc.VarType|str|None): The data type of variable.
            Default: None
156
        lod_level (int|None): The level of lod tensor. 0 means it is not a time
157
            series data.
158
            Default: None
159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180
        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')
181 182
    """

Y
Yu Yang 已提交
183 184
    def __init__(self,
                 block,
Y
Yu Yang 已提交
185
                 type=core.VarDesc.VarType.LOD_TENSOR,
Y
Yu Yang 已提交
186 187 188 189
                 name=None,
                 shape=None,
                 dtype=None,
                 lod_level=None,
190
                 capacity=None,
Q
QI JUN 已提交
191
                 persistable=None,
F
fengjiayi 已提交
192
                 error_clip=None,
Y
Yu Yang 已提交
193
                 stop_gradient=False,
F
fengjiayi 已提交
194
                 is_data=False,
Y
Yu Yang 已提交
195
                 **kwargs):
Y
Yu Yang 已提交
196
        self.block = block
F
fengjiayi 已提交
197
        self.error_clip = error_clip
Y
Yu Yang 已提交
198 199

        if name is None:
Y
Yu Yang 已提交
200
            name = unique_name.generate('_generated_var')
D
Dong Zhihong 已提交
201
        is_new_var = False
M
minqiyang 已提交
202 203
        name = cpt.to_literal_str(name)
        self.desc = self.block.desc.find_var(cpt.to_bytes(name))
D
Dong Zhihong 已提交
204 205

        if self.desc is None:
M
minqiyang 已提交
206
            self.desc = self.block.desc.var(cpt.to_bytes(name))
Y
Yu Yang 已提交
207
            is_new_var = True
Y
Yu Yang 已提交
208

Y
Yu Yang 已提交
209 210 211 212 213 214 215 216
        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
217
        if shape is not None:
Y
Yu Yang 已提交
218
            if is_new_var:
219
                self.desc.set_shape(shape)
Y
Yu Yang 已提交
220 221 222 223 224 225 226 227
            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
228
        if dtype is not None:
229
            if not isinstance(dtype, core.VarDesc.VarType):
230
                dtype = convert_np_dtype_to_dtype_(dtype)
Y
Yu Yang 已提交
231
            if is_new_var:
F
fengjiayi 已提交
232
                self.desc.set_dtype(dtype)
Y
Yu Yang 已提交
233
            else:
F
fengjiayi 已提交
234
                old_dtype = self.dtype
Q
QI JUN 已提交
235
                if dtype != old_dtype:
Y
Yu Yang 已提交
236 237 238 239 240
                    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
241 242

        if lod_level is not None:
Y
Yu Yang 已提交
243
            if is_new_var:
244
                self.desc.set_lod_level(lod_level)
Y
Yu Yang 已提交
245 246 247 248 249 250 251
            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))
252 253 254 255 256 257 258 259 260 261 262
        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))

263 264 265 266 267 268 269 270
        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
271
        self.block.vars[name] = self
Y
Yu Yang 已提交
272
        self.op = None
Y
Yu Yang 已提交
273
        self.stop_gradient = stop_gradient
F
fengjiayi 已提交
274
        self.is_data = is_data
Y
Yu Yang 已提交
275

276
    def __str__(self):
Y
Yang Yang(Tony) 已提交
277 278
        return self.to_string(True)

F
update  
fengjiayi 已提交
279
    def to_string(self, throw_on_error, with_details=False):
280 281 282 283
        """
        Get debug string.

        Args:
284 285
            throw_on_error(bool): True if raise an exception when self is
                not initialized.
F
update  
fengjiayi 已提交
286
            with_details(bool): more details about variables and parameters
287 288
                (e.g. trainable, optimize_attr, ...) will be printed when
                with_details is True. Default False;
289

290 291
        Returns:
            str: The debug string.
292
        """
F
update  
fengjiayi 已提交
293 294
        assert isinstance(throw_on_error, bool) and isinstance(with_details,
                                                               bool)
295
        protostr = self.desc.serialize_to_string()
296
        proto = framework_pb2.VarDesc.FromString(six.binary_type(protostr))
F
update  
fengjiayi 已提交
297 298 299 300
        res_str = _debug_string_(proto, throw_on_error)
        if with_details:
            additional_attr = ("error_clip", "stop_gradient")
            for attr_name in additional_attr:
301 302
                res_str += "%s: %s\n" % (
                    attr_name, six.binary_type(getattr(self, attr_name)))
F
update  
fengjiayi 已提交
303
        return res_str
304 305 306

    __repr__ = __str__

W
Wu Yi 已提交
307
    def _set_desc(self, input):
308 309 310 311 312 313 314 315 316
        """
        Set the variable description.

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

        Returns:
            None
        """
317 318
        self.desc = input

319 320 321 322
    @property
    def persistable(self):
        return self.desc.persistable()

Y
Yu Yang 已提交
323 324 325 326
    @persistable.setter
    def persistable(self, p):
        self.desc.set_persistable(p)

Y
Yu Yang 已提交
327 328
    @property
    def name(self):
M
minqiyang 已提交
329
        return cpt.to_literal_str(self.desc.name())
Y
Yu Yang 已提交
330

T
typhoonzero 已提交
331 332 333 334
    @name.setter
    def name(self, new_name):
        self.desc.set_name(new_name)

Y
Yu Yang 已提交
335 336 337
    @property
    def shape(self):
        # convert to tuple, make it as same as numpy API.
338
        return tuple(self.desc.shape())
Y
Yu Yang 已提交
339 340

    @property
F
fengjiayi 已提交
341 342
    def dtype(self):
        return self.desc.dtype()
Y
Yu Yang 已提交
343 344 345

    @property
    def lod_level(self):
346
        return self.desc.lod_level()
Y
Yu Yang 已提交
347

Y
Yu Yang 已提交
348 349 350 351
    @property
    def type(self):
        return self.desc.type()

W
Wu Yi 已提交
352
    def _set_error_clip(self, error_clip):
353 354 355 356 357 358 359 360 361
        """
        Set the error_clip.

        Args:
            error_clip(BaseErrorClipAttr) : The new error_clip.

        Returns:
            None
        """
362 363
        self.error_clip = error_clip

Y
Yu Yang 已提交
364

F
fengjiayi 已提交
365 366 367
def get_all_op_protos():
    """
    Get all registered op proto from PaddlePaddle C++ end.
368

369 370
    Returns:
       list: list of OpProto.
F
fengjiayi 已提交
371 372 373 374
    """
    protostrs = core.get_all_op_protos()
    ret_values = []
    for pbstr in protostrs:
375
        op_proto = framework_pb2.OpProto.FromString(six.binary_type(pbstr))
F
fengjiayi 已提交
376 377 378 379 380
        ret_values.append(op_proto)
    return ret_values


class OpProtoHolder(object):
381 382 383 384
    """
    A global variable to hold all OpProtos from C++ as a map
    """

F
fengjiayi 已提交
385 386 387 388 389 390 391 392 393
    @classmethod
    def instance(cls):
        if not hasattr(cls, '_instance'):
            cls._instance = cls()
        return cls._instance

    def __init__(self):
        assert not hasattr(
            self.__class__,
394
            '_instance'), 'Please use `instance()` to get OpProtoHolder object!'
F
fengjiayi 已提交
395 396 397 398 399 400
        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):
401 402 403 404 405 406 407 408
        """
        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
409 410
        if type not in self.op_proto_map:
            raise ValueError("Operator \"%s\" has not been registered." % type)
F
fengjiayi 已提交
411 412
        return self.op_proto_map[type]

413 414 415 416 417 418 419
    @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
420

Y
Yu Yang 已提交
421
class Operator(object):
422
    """
423 424 425 426 427 428 429
    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
430
        type(str): The type of operator. Default None.
431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450
        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
451
        Block.append_op or Block._prepend_op instead.
452 453 454 455 456 457 458 459 460 461

    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]})
462
    """
463 464 465 466 467
    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
468
        'channel_recv', 'select', 'checkpoint_notify', 'gen_nccl_id'
469
    }
470

Y
Yu Yang 已提交
471 472
    def __init__(self,
                 block,
Y
Yu Yang 已提交
473
                 desc,
Y
Yu Yang 已提交
474 475 476 477 478
                 type=None,
                 inputs=None,
                 outputs=None,
                 attrs=None):
        self.block = block
Y
Yu Yang 已提交
479
        self.desc = desc
T
typhoonzero 已提交
480
        self.attrs = attrs
Y
yuyang18 已提交
481 482 483 484 485 486 487 488
        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
489 490 491 492 493 494 495 496

        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
497

F
fengjiayi 已提交
498 499 500 501 502
        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
503
        self.desc.set_type(type)
F
fengjiayi 已提交
504
        proto = OpProtoHolder.instance().get_op_proto(type)
505

Y
Yang Yang(Tony) 已提交
506 507
        def find_name(var_list, name):
            for var_name in var_list:
Q
Qiao Longfei 已提交
508
                if var_list[var_name] is not None and var_name == name:
Y
Yang Yang(Tony) 已提交
509 510
                    return True
            return False
Q
QI JUN 已提交
511

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

Y
Yu Yang 已提交
538
        if outputs is not None:
539 540 541 542 543 544 545
            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
546 547
                raise ValueError(("Incorrect setting for output(s) of "
                                  "operator \"%s\". Need: [%s] Given: [%s]") %
548 549 550
                                 (type,
                                  ", ".join(six.binary_type(e) for e in need),
                                  ", ".join(six.binary_type(e) for e in given)))
551

F
fengjiayi 已提交
552
            for out_proto in proto.outputs:
553 554 555 556
                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
557 558
                    raise ValueError(
                        "Output %s expects only one output, but %d are given." %
559 560 561
                        (out_proto.name, len(out_args)))
                out_arg_names = []
                for arg in out_args:
M
minqiyang 已提交
562
                    out_arg_names.append(cpt.to_literal_str(arg.name))
563 564
                    arg.op = self
                self.desc.set_output(out_proto.name, out_arg_names)
F
Update  
fengjiayi 已提交
565

Y
yuyang18 已提交
566 567
        if self.attrs is not None:
            if not isinstance(self.attrs, dict):
568
                raise TypeError("'attrs' should be a dict.")
F
fengjiayi 已提交
569
            for attr in proto.attrs:
F
Update  
fengjiayi 已提交
570
                attr_name = attr.name
Y
yuyang18 已提交
571 572
                if (attr_name not in self.attrs) or (
                        self.attrs[attr_name] is None):
F
Update  
fengjiayi 已提交
573
                    continue
Y
Yancey1989 已提交
574
                attr_val = self.attrs[attr_name]
G
gongweibao 已提交
575 576
                self._update_desc_attr(attr_name, attr_val)

577
        self.desc.check_attrs()
578
        if self.has_kernel(type):
Q
QI JUN 已提交
579
            self.desc.infer_var_type(self.block.desc)
Y
Yu Yang 已提交
580
            self.desc.infer_shape(self.block.desc)
F
fengjiayi 已提交
581

582 583 584
    def has_kernel(self, op_type):
        return op_type not in self.OP_WITHOUT_KERNEL_SET

Y
Yang Yang(Tony) 已提交
585
    def to_string(self, throw_on_error):
586
        """
587 588
        Get debug string.

589
        Args:
590 591
            throw_on_error(bool): Whether to raise exception if self is not
                initialized.
592

593 594
        Returns:
            str: The debug string.
595 596

        """
597
        protostr = self.desc.serialize_to_string()
598
        proto = framework_pb2.OpDesc.FromString(six.binary_type(protostr))
Y
Yang Yang(Tony) 已提交
599 600 601 602
        return _debug_string_(proto, throw_on_error)

    def __str__(self):
        return self.to_string(True)
603 604 605

    __repr__ = __str__

F
fengjiayi 已提交
606 607 608 609 610
    @property
    def type(self):
        return self.desc.type()

    def input(self, name):
611
        """
612
        Get the input arguments according to the input parameter name.
613

614 615
        Args:
            name(str): The input parameter name.
616

617 618 619
        Returns:
            list: return the list of argument names that associated with \
                the specific parameter name.
620
        """
F
fengjiayi 已提交
621 622
        return self.desc.input(name)

T
typhoonzero 已提交
623
    def rename_input(self, old_name, new_name):
624 625 626 627 628 629 630 631 632 633
        """
        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
634 635 636
        self.desc.rename_input(old_name, new_name)

    def rename_output(self, old_name, new_name):
637 638 639 640 641 642 643 644 645 646
        """
        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
647 648
        self.desc.rename_output(old_name, new_name)

F
fengjiayi 已提交
649 650 651 652
    @property
    def input_names(self):
        return self.desc.input_names()

T
typhoonzero 已提交
653 654 655 656 657 658 659 660
    @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
661
    def output(self, name):
662
        """
663
        Get output arguments by the output parameter name.
664

665 666
        Args:
            name(str): The output parameter name.
667

668 669 670
        Returns:
            list: return the list of argument names associated with \
                the specific parameter name.
671
        """
F
fengjiayi 已提交
672 673 674 675 676 677
        return self.desc.output(name)

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

678 679 680 681 682 683 684 685
    @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
686
    def has_attr(self, name):
687
        """
688 689
        Whether this Operator has the attribute with name or not.

690
        Args:
691
            name(str): the attribute name.
692

693 694
        Returns:
            bool: True if has this attribute.
695 696

        """
F
fengjiayi 已提交
697 698 699
        return self.desc.has_attr(name)

    def attr_type(self, name):
700
        """
701
        Get the type of attribute by attribute's name.
702

703 704
        Args:
            name(str): the attribute name.
705

706 707
        Returns:
            core.AttrType: the attribute type.
708
        """
F
fengjiayi 已提交
709 710
        return self.desc.attr_type(name)

Y
yuyang18 已提交
711
    def set_attr(self, name, val):
712 713 714 715 716 717 718 719 720 721
        """
        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
722
        self.attrs[name] = val
G
gongweibao 已提交
723 724 725 726 727 728 729 730 731 732 733 734 735
        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
736 737
        if isinstance(val, Block):
            self.desc.set_block_attr(name, val.desc)
Y
Yancey1989 已提交
738 739
        elif isinstance(val, list) and val and all(
                isinstance(v, Block) for v in val):
740
            self.desc.set_blocks_attr(name, [v.desc for v in val])
Q
Qiyang Min 已提交
741 742 743 744 745
        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
746

F
fengjiayi 已提交
747 748 749 750 751
    @property
    def attr_names(self):
        return self.desc.attr_names()

    def attr(self, name):
752
        """
753 754
        Get the attribute by name.

755
        Args:
756
            name(str): the attribute name.
757

758 759
        Returns:
            bool|int|str|float|list: The attribute value. The return value
760 761
            can be any valid attribute type.
        """
F
fengjiayi 已提交
762
        return self.desc.attr(name)
Y
Yu Yang 已提交
763

F
fengjiayi 已提交
764
    def block_attr(self, name):
765
        """
766
        Get the block attribute by name.
767

768 769
        Args:
            name(str): the attribute name.
770

771 772
        Returns:
            int: the block index.
773
        """
F
fengjiayi 已提交
774
        return self.desc.block_attr(name)
Y
Yu Yang 已提交
775

J
JiayiFeng 已提交
776
    def all_attrs(self):
F
fengjiayi 已提交
777
        """
778 779 780 781
        Get the attribute dict.

        Returns:
            dict: The Operator's attribute dict.
F
fengjiayi 已提交
782 783 784 785 786 787 788 789 790 791
        """
        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
792

Y
Yu Yang 已提交
793
class Block(object):
794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822
    """
    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
823
    def __init__(self, program, idx):
Y
Yu Yang 已提交
824
        self.desc = program.desc.block(idx)
825
        self.vars = collections.OrderedDict()  # var_name --> var
Q
qiaolongfei 已提交
826
        self.ops = list()  # operator list
Y
Yu Yang 已提交
827
        self.program = program
828
        self.removed_vars = collections.OrderedDict()
Y
Yu Yang 已提交
829

830
    def __str__(self):
Y
Yang Yang(Tony) 已提交
831 832
        return self.to_string(True)

F
fengjiayi 已提交
833 834
    def to_string(self, throw_on_error, with_details=False):
        """
835 836
        Get debug string.

F
fengjiayi 已提交
837 838
        Args:
            throw_on_error(bool): raise exception when self is not initialized
839
                when throw_on_error is True.
F
update  
fengjiayi 已提交
840
            with_details(bool): more details about variables and parameters
841 842
                (e.g. trainable, optimize_attr, ...) will be printed when
                with_details is True. Default False.
F
fengjiayi 已提交
843

844 845
        Returns:
            str: The debug string.
F
fengjiayi 已提交
846 847 848 849
        """
        assert isinstance(throw_on_error, bool) and isinstance(with_details,
                                                               bool)
        if with_details:
F
fengjiayi 已提交
850
            re_add_indent = re.compile(r"\n(.)")
F
fengjiayi 已提交
851 852
            res_str = "blocks {\n  idx: %d\n  parent_idx: %d" % (
                self.idx, self.parent_idx)
853
            for var in list(self.vars.values()):
F
fengjiayi 已提交
854
                res_str += "\n  vars {\n    %s  }" % re_add_indent.sub(
F
update  
fengjiayi 已提交
855
                    r"\n    \1", var.to_string(throw_on_error, with_details))
F
fengjiayi 已提交
856
            for op in self.ops:
F
fengjiayi 已提交
857 858
                res_str += "\n  ops {\n    %s  }" % re_add_indent.sub(
                    r"\n    \1", op.to_string(throw_on_error))
F
fengjiayi 已提交
859 860 861
            res_str += "\n}"
        else:
            protostr = self.desc.serialize_to_string()
862 863
            proto = framework_pb2.BlockDesc.FromString(
                six.binary_type(protostr))
F
fengjiayi 已提交
864 865
            res_str = _debug_string_(proto, throw_on_error)
        return res_str
866 867 868

    __repr__ = __str__

Y
Yu Yang 已提交
869 870
    @property
    def parent_idx(self):
Y
Yu Yang 已提交
871
        return self.desc.parent
Y
Yu Yang 已提交
872

Y
Yu Yang 已提交
873 874 875 876
    @property
    def forward_block_idx(self):
        return self.desc.get_forward_block_idx()

W
Wu Yi 已提交
877
    def _set_forward_block_idx(self, idx):
878 879 880 881 882 883 884 885 886
        """
        Set the forward block Idx.

        Args:
            idx(int): the block index.

        Returns:
            None
        """
W
Wu Yi 已提交
887
        self.desc._set_forward_block_idx(idx)
Y
Yu Yang 已提交
888

Y
Yu Yang 已提交
889 890
    @property
    def idx(self):
Y
Yu Yang 已提交
891
        return self.desc.id
Y
Yu Yang 已提交
892

Q
Qiao Longfei 已提交
893
    def var(self, name):
894 895 896 897 898 899 900 901 902 903 904 905 906
        """
        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.
        """
907
        if not isinstance(name, six.string_types):
908 909 910 911
            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
912 913
        v = self.vars.get(name, None)
        if v is None:
Q
Qiao Longfei 已提交
914
            raise ValueError("var %s not in this block" % name)
Y
Yu Yang 已提交
915
        return v
Q
Qiao Longfei 已提交
916

W
Wu Yi 已提交
917
    def _var_recursive(self, name):
918 919 920 921 922 923 924 925 926 927 928 929 930
        """
        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
931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956
        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
957

Q
Qiao Longfei 已提交
958
    def all_parameters(self):
959
        return list(self.iter_parameters())
960

961
    def iter_parameters(self):
962
        return (item[1] for item in list(self.vars.items())
963
                if isinstance(item[1], Parameter))
Q
Qiao Longfei 已提交
964

Y
Yu Yang 已提交
965
    def create_var(self, *args, **kwargs):
966
        var = Variable(block=self, *args, **kwargs)
967 968
        if 'initializer' in kwargs:
            kwargs['initializer'](var, self)
Q
Qiao Longfei 已提交
969
        return var
Y
Yu Yang 已提交
970

Q
Qiao Longfei 已提交
971 972 973
    def has_var(self, name):
        return name in self.vars

W
Wu Yi 已提交
974
    def _rename_var(self, name, new_name):
T
typhoonzero 已提交
975 976
        """
        Rename variable in vars and ops' inputs and outputs
977 978 979 980 981 982 983 984 985 986 987 988

        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
989
        """
M
minqiyang 已提交
990 991 992
        name = cpt.to_literal_str(name)
        new_name = cpt.to_literal_str(new_name)

T
typhoonzero 已提交
993
        if not self.has_var(name):
994
            raise ValueError("var %s is not in current block" % name)
T
wip  
typhoonzero 已提交
995 996
        v = self.var(name)
        if type(v) == Parameter:
T
typhoonzero 已提交
997
            var_type = "Parameter"
T
wip  
typhoonzero 已提交
998 999 1000 1001 1002 1003 1004
            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
1005
            var_type = "Variable"
T
wip  
typhoonzero 已提交
1006 1007 1008 1009
            error_clip = v.error_clip
            stop_gradient = v.stop_gradient
        else:
            raise ValueError("unsupported var type: %s", type(v))
T
typhoonzero 已提交
1010
        orig_var_type = v.type
M
minqiyang 已提交
1011
        self.desc._rename_var(cpt.to_bytes(name), cpt.to_bytes(new_name))
W
Wu Yi 已提交
1012
        # NOTE: v is destroyed by C++ after calling _rename_var.
M
minqiyang 已提交
1013
        d = self.desc.find_var(cpt.to_bytes(new_name))
T
typhoonzero 已提交
1014
        if var_type == "Parameter":
T
wip  
typhoonzero 已提交
1015 1016 1017 1018
            var = Parameter(
                self,
                d.shape(),
                d.dtype(),
T
typhoonzero 已提交
1019
                type=orig_var_type,
T
wip  
typhoonzero 已提交
1020 1021 1022 1023 1024 1025 1026
                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
1027
        elif var_type == "Variable":
T
wip  
typhoonzero 已提交
1028 1029
            var = Variable(
                self,
T
typhoonzero 已提交
1030
                type=orig_var_type,
T
wip  
typhoonzero 已提交
1031 1032 1033 1034
                name=new_name,
                error_clip=error_clip,
                stop_gradient=stop_gradient)

W
Wu Yi 已提交
1035
        # rename the python side, _sync_with_cpp will only add
T
wip  
typhoonzero 已提交
1036 1037 1038
        # new vars/ops to python side.
        self.vars[new_name] = var
        del self.vars[name]
W
Wu Yi 已提交
1039
        self._sync_with_cpp()
1040
        return var
T
typhoonzero 已提交
1041

W
Wu Yi 已提交
1042 1043
    def _remove_var(self, name):
        self._sync_with_cpp()
M
minqiyang 已提交
1044
        self.desc._remove_var(cpt.to_bytes(name))
1045 1046
        del self.vars[name]

Y
Yu Yang 已提交
1047 1048
    def create_parameter(self, *args, **kwargs):
        global_block = self.program.global_block()
Q
Qiao Longfei 已提交
1049
        param = Parameter(global_block, *args, **kwargs)
1050 1051
        if 'initializer' in kwargs:
            kwargs['initializer'](param, self)
Q
Qiao Longfei 已提交
1052
        return param
Y
Yu Yang 已提交
1053

Y
Yu Yang 已提交
1054
    def append_op(self, *args, **kwargs):
1055 1056 1057 1058 1059 1060
        """
        Appends a new Operator according to the giving arguments.

        Returns:
            Operator: the append Operator.
        """
Y
Yu Yang 已提交
1061
        op_desc = self.desc.append_op()
1062
        op = Operator(block=self, desc=op_desc, *args, **kwargs)
Y
Yu Yang 已提交
1063 1064 1065
        self.ops.append(op)
        return op

W
Wu Yi 已提交
1066
    def _insert_op(self, index, *args, **kwargs):
1067 1068 1069 1070 1071 1072 1073 1074 1075
        """
        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
1076 1077
        self._sync_with_cpp()
        op_desc = self.desc._insert_op(index)
Q
qiaolongfei 已提交
1078 1079 1080 1081
        op = Operator(block=self, desc=op_desc, *args, **kwargs)
        self.ops.insert(index, op)
        return op

W
Wu Yi 已提交
1082
    def _remove_op(self, index):
1083 1084 1085 1086 1087 1088 1089 1090 1091
        """
        Remove the specific position operator.

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

        Returns:
            None
        """
W
Wu Yi 已提交
1092 1093
        self._sync_with_cpp()
        self.desc._remove_op(index, index + 1)
1094 1095
        del self.ops[index]

W
Wu Yi 已提交
1096
    def _slice_ops(self, start, end):
1097 1098 1099 1100 1101 1102 1103 1104 1105 1106
        """
        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
1107
        return self.ops[start:end]
Y
Yancey1989 已提交
1108

W
Wu Yi 已提交
1109 1110
    def _prepend_op(self, *args, **kwargs):
        op_desc = self.desc._prepend_op()
Y
Yu Yang 已提交
1111
        op = Operator(self, op_desc, *args, **kwargs)
Q
qiaolongfei 已提交
1112
        self.ops.insert(0, op)
Y
Yu Yang 已提交
1113 1114
        return op

W
Wu Yi 已提交
1115
    def _sync_with_cpp(self):
1116
        """
1117 1118
        Sync from the desc on the c++ end. This method is used to synchronize
        the c++ desc instance generated by backward.
1119
        """
Q
Qiao Longfei 已提交
1120 1121 1122 1123 1124
        # 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())

1125
        # sync variables removed from c++ end
1126
        for var in list(self.vars.keys()):
M
minqiyang 已提交
1127
            if not self.desc.find_var(cpt.to_bytes(var)):
1128 1129
                self.vars.pop(var)

Q
Qiao Longfei 已提交
1130
        # sync operators from cpp
1131 1132 1133 1134
        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
1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150
        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
1151 1152 1153 1154 1155

        # 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
1156
            self.ops.insert(0, op)
Q
Qiao Longfei 已提交
1157 1158 1159 1160 1161 1162 1163

        # 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)

1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176
        # 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
1177 1178 1179 1180
        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
1181
    def _copy_param_info_from(self, other):
1182
        """
1183 1184
        Copy the information of parameters from the other block.

1185
        Args:
1186 1187 1188 1189 1190
            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.
1191 1192 1193 1194 1195

        Returns:
            None
        """
        if not isinstance(other, Block):
W
Wu Yi 已提交
1196 1197
            raise TypeError(
                "_copy_param_info_from should be invoked with Block")
1198
        for p in other.iter_parameters():
1199 1200 1201
            assert isinstance(p, Parameter)
            v = self.vars.get(p.name, None)
            if v is None:
W
Wu Yi 已提交
1202
                raise ValueError("_copy_param_info_from should be invoked with "
1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214
                                 "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
1215
                gradient_clip_attr=p.gradient_clip_attr,
F
fengjiayi 已提交
1216
                error_clip=p.error_clip,
1217 1218 1219
                name=v.name)
            self.vars[new_p.name] = new_p

W
Wu Yi 已提交
1220
    def _clone_variable(self, var):
1221 1222
        """
        Clone a variable into current block.
1223

1224 1225 1226 1227
        Args:
            var: the variable to be cloned.

        Returns:
1228
            Variable: the new  variable cloned from 'var' in current block.
1229 1230
        """
        assert isinstance(var, Variable)
T
update  
typhoonzero 已提交
1231 1232 1233 1234 1235
        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
1236 1237
        elif var.type == core.VarDesc.VarType.RAW:
            ret_var = self.create_var(
T
tangwei12 已提交
1238
                name=var.name, persistable=var.persistable, type=var.type)
T
typhoonzero 已提交
1239 1240 1241 1242 1243 1244
        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
1245 1246
                persistable=True,
                is_data=var.is_data)
T
update  
typhoonzero 已提交
1247 1248 1249 1250 1251 1252 1253
        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
1254 1255
                persistable=True,
                is_data=var.is_data)
T
update  
typhoonzero 已提交
1256
        return ret_var
1257

Y
Yu Yang 已提交
1258 1259

class Program(object):
D
dzhwinter 已提交
1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270
    """
    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
1271
    default_main_program run in every mini batch and adjust the weights.
D
dzhwinter 已提交
1272 1273

    Returns:
Y
yuyang18 已提交
1274
        A empty program.
D
dzhwinter 已提交
1275 1276

    Examples:
Y
yuyang18 已提交
1277 1278 1279 1280 1281 1282
        >>> 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
1283 1284 1285

    """

1286 1287
    def __init__(self):
        self.desc = core.ProgramDesc()
Y
Yu Yang 已提交
1288 1289
        self.blocks = [Block(self, 0)]
        self.current_block_idx = 0
D
dzhwinter 已提交
1290
        self._seed = 0
Y
yuyang18 已提交
1291
        self._current_role = core.op_proto_and_checker_maker.OpRole.Forward
Y
yuyang18 已提交
1292
        self._op_role_var = []
Y
yuyang18 已提交
1293 1294 1295

    @property
    def op_role(self):
Y
yuyang18 已提交
1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308
        """
        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
1309 1310 1311 1312 1313 1314 1315 1316
        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
1317 1318 1319 1320 1321 1322 1323
        """
        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
1324 1325 1326 1327
        return self._op_role_var

    @op_role_var.setter
    def set_op_role_var(self, var_name):
Y
yuyang18 已提交
1328
        self._op_role_var = [var_name]
Y
yuyang18 已提交
1329 1330

    @contextlib.contextmanager
1331
    def optimized_guard(self, param_and_grads):
Y
yuyang18 已提交
1332 1333 1334 1335 1336 1337 1338
        """
        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:
1339
            param_and_grads(list): The variables (names) to be optimized.
Y
yuyang18 已提交
1340 1341 1342 1343

        Examples:

            >>> p, g = backward(...)
1344
            >>> with program.optimized_guard([p,g]):
Y
yuyang18 已提交
1345 1346
            >>>     p = p - 0.001 * g
        """
Y
yuyang18 已提交
1347 1348
        OpRole = core.op_proto_and_checker_maker.OpRole
        self._current_role = OpRole.Optimize
1349 1350 1351 1352
        self._op_role_var = [
            var.name if isinstance(var, Variable) else var
            for var in param_and_grads
        ]
Y
yuyang18 已提交
1353
        yield
Y
yuyang18 已提交
1354
        self._op_role_var = []
Y
yuyang18 已提交
1355
        self._current_role = OpRole.Forward
Y
Yu Yang 已提交
1356

1357
    def __str__(self):
Y
yuyang18 已提交
1358 1359 1360 1361 1362 1363 1364 1365 1366
        """
        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
1367 1368
        return self.to_string(True)

F
fengjiayi 已提交
1369 1370 1371
    def to_string(self, throw_on_error, with_details=False):
        """
        To debug string.
Y
yuyang18 已提交
1372

F
fengjiayi 已提交
1373
        Args:
Y
yuyang18 已提交
1374 1375
            throw_on_error(bool): raise Value error when any of required fields
                is not set.
F
fengjiayi 已提交
1376

Y
yuyang18 已提交
1377 1378 1379 1380 1381 1382 1383 1384 1385 1386
            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
1387 1388 1389 1390 1391 1392 1393 1394 1395 1396

        """
        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()
1397 1398
            proto = framework_pb2.ProgramDesc.FromString(
                six.binary_type(protostr))
F
fengjiayi 已提交
1399 1400
            res_str = _debug_string_(proto, throw_on_error)
        return res_str
1401

1402
    def get_desc(self):
Y
yuyang18 已提交
1403 1404 1405 1406 1407 1408 1409
        """
        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.
        """
1410 1411
        return self.desc

1412
    def clone(self, for_test=False):
Y
yuyang18 已提交
1413 1414 1415
        """
        Create a new, duplicated program.

1416

Y
yuyang18 已提交
1417 1418 1419 1420
        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`.
1421

Y
yuyang18 已提交
1422 1423 1424 1425
        * 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
1426 1427 1428 1429 1430
        :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()
1431 1432

        Args:
Y
yuyang18 已提交
1433 1434
            for_test(bool): True if change the :code:`is_test` attribute of
                operators to :code:`True`.
1435

D
dzhwinter 已提交
1436
        Returns:
Y
yuyang18 已提交
1437 1438 1439 1440 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
            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.
1490 1491
        """
        if for_test:
1492
            p = self.inference_optimize()
1493
        else:
1494
            p = Program()
1495
            p.desc = core.ProgramDesc(self.desc)
M
minqiyang 已提交
1496 1497 1498
            p.blocks = [
                Block(p, i) for i in six.moves.range(self.desc.num_blocks())
            ]
W
Wu Yi 已提交
1499
            p._sync_with_cpp()
1500

W
Wu Yi 已提交
1501
        p._copy_param_info_from(self)
F
fengjiayi 已提交
1502
        p.copy_data_info_from(self)
Y
Yu Yang 已提交
1503
        return p
1504

1505
    def prune(self, targets):
Y
yuyang18 已提交
1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520
        """
        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.

        """
1521 1522 1523 1524 1525 1526
        if not isinstance(targets, list):
            targets = [targets]
        targets_idx = []
        for t in targets:
            if not isinstance(t, Operator):
                if isinstance(t, Variable):
1527 1528
                    # After transpiler processing, the op that output this
                    # variable maybe has been changed, so t.op is not reliable
1529
                    # and we need to find the current op that generate this
1530 1531 1532 1533 1534 1535 1536 1537
                    # 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

1538
                    t = t.op
1539 1540 1541 1542
                    if t is None:
                        raise ValueError(
                            "The target variable must have an "
                            "associated operator that generates it.")
1543
                else:
1544 1545
                    raise ValueError("All targets of prune() can only be "
                                     "Variable or Operator.")
1546 1547 1548 1549

            targets_idx.append([t.block.idx, t.idx])
        res = Program()
        res.desc = core.prune(self.desc, targets_idx)
M
minqiyang 已提交
1550 1551 1552
        res.blocks = [
            Block(res, i) for i in six.moves.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
        """
F
fengjiayi 已提交
1558 1559 1560 1561 1562
        This method will create a new program and do following adjustments on it:
        1. Remove all reader variables and their creator ops if exist.

        2. Remove the :code:`read_op` if exists.

1563
        3. change the :code:`is_test`
Y
yuyang18 已提交
1564 1565 1566 1567 1568 1569 1570 1571 1572
        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.
        """
1573 1574
        # this is an alternative implement before
        # core.inference_optimize being fixed.
1575
        res = Program()
1576
        res.desc = core.ProgramDesc(self.desc)
F
fengjiayi 已提交
1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592

        # remove all readers and the read_op if exist
        read_op_idx = 0
        root_block = res.desc.block(0)
        while True:
            if read_op_idx >= root_block.op_size() or root_block.op(
                    read_op_idx).type() == 'read':
                break
            read_op_idx += 1
        if read_op_idx < root_block.op_size():
            root_block._remove_op(0, read_op_idx + 1)
        for var in root_block.all_vars():
            if var.type() == core.VarDesc.VarType.READER:
                root_block._remove_var(var.name())

        # change all `is_test` attributes to True
M
minqiyang 已提交
1593
        for i in six.moves.range(res.desc.num_blocks()):
1594
            block = res.desc.block(i)
M
minqiyang 已提交
1595
            for j in six.moves.range(block.op_size()):
1596 1597 1598
                op = block.op(j)
                if op.has_attr('is_test'):
                    op.set_attr('is_test', True)
M
minqiyang 已提交
1599 1600 1601
        res.blocks = [
            Block(res, i) for i in six.moves.range(res.desc.num_blocks())
        ]
W
Wu Yi 已提交
1602
        res._sync_with_cpp()
1603 1604
        return res

1605 1606
    @staticmethod
    def parse_from_string(binary_str):
Y
yuyang18 已提交
1607 1608 1609 1610 1611 1612 1613
        """
        Deserialize a program desc from protobuf binary string.

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

        Args:
1614
            binary_str_type(str): The binary prootbuf string.
Y
yuyang18 已提交
1615 1616 1617 1618

        Returns:
            Program: A deserialized program desc.
        """
1619 1620
        p = Program()
        p.desc = core.ProgramDesc(binary_str)
M
minqiyang 已提交
1621
        p.blocks = [Block(p, i) for i in six.moves.range(p.desc.num_blocks())]
W
Wu Yi 已提交
1622
        p._sync_with_cpp()
1623
        return p
Y
Yu Yang 已提交
1624

D
dzhwinter 已提交
1625 1626
    @property
    def random_seed(self):
Y
yuyang18 已提交
1627 1628 1629 1630 1631 1632
        """
        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
1633 1634
        return self._seed

Q
qiaolongfei 已提交
1635 1636
    @property
    def num_blocks(self):
Y
yuyang18 已提交
1637 1638 1639
        """
        The number of blocks in this program.
        """
Q
qiaolongfei 已提交
1640 1641
        return self.desc.num_blocks()

D
dzhwinter 已提交
1642 1643 1644 1645 1646 1647
    @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
1648
    def __repr__(self):
1649
        return self.__str__()
1650

Y
Yu Yang 已提交
1651
    def global_block(self):
Y
yuyang18 已提交
1652 1653 1654
        """
        Get the first block of this program.
        """
Y
Yu Yang 已提交
1655 1656
        return self.blocks[0]

Q
Qiao Longfei 已提交
1657
    def block(self, index):
Y
yuyang18 已提交
1658 1659 1660 1661 1662 1663 1664 1665
        """
        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
1666 1667
        return self.blocks[index]

Y
Yu Yang 已提交
1668
    def current_block(self):
Y
yuyang18 已提交
1669 1670 1671 1672
        """
        Get the current block. The :code:`current` block is the block to append
        operators.
        """
Y
Yu Yang 已提交
1673 1674
        return self.blocks[self.current_block_idx]

F
update  
fengjiayi 已提交
1675
    def create_block(self, parent_idx=None):
Y
yuyang18 已提交
1676 1677 1678 1679 1680 1681 1682 1683 1684 1685
        """
        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
1686
        new_block_idx = len(self.blocks)
F
update  
fengjiayi 已提交
1687 1688 1689
        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
1690 1691 1692 1693 1694
        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
1695 1696 1697 1698 1699
        """
        Exit a code block, i.e., roll back to the parent block.
        Returns:
            None
        """
Y
Yu Yang 已提交
1700 1701
        self.current_block_idx = self.current_block().parent_idx

W
Wu Yi 已提交
1702
    def _sync_with_cpp(self):
Y
yuyang18 已提交
1703 1704 1705 1706 1707 1708 1709 1710 1711 1712
        """
        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
1713 1714 1715
        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
1716
            block._sync_with_cpp()
Q
Qiao Longfei 已提交
1717

W
Wu Yi 已提交
1718
    def _copy_param_info_from(self, other):
1719
        """
1720
        Copy the information of parameters from other program.
D
dzhwinter 已提交
1721

Y
yuyang18 已提交
1722 1723 1724
        Notes: This is a very low level API. Users should not invoke it
        directly.

1725 1726 1727 1728 1729 1730 1731
        Args:
            other(Program): Other program

        Returns:
            None
        """
        if not isinstance(other, Program):
W
Wu Yi 已提交
1732
            raise TypeError("_copy_param_info_from should be invoked with "
1733 1734 1735
                            "Program")

        if len(self.blocks) != len(other.blocks):
W
Wu Yi 已提交
1736
            raise ValueError("_copy_param_info_from should be invoked with two "
1737
                             "program, with represent the same topology")
W
Wu Yi 已提交
1738
        self.global_block()._copy_param_info_from(other.global_block())
1739

F
fengjiayi 已提交
1740 1741 1742
    def copy_data_info_from(self, other):
        """
        Copy the information of data variables from other program.
D
dzhwinter 已提交
1743

Y
yuyang18 已提交
1744 1745 1746
        Notes: This is a very low level API. Users should not invoke it
        directly.

F
fengjiayi 已提交
1747 1748 1749 1750 1751 1752 1753
        Args:
            other(Program): Other program

        Returns:
            None
        """
        if not isinstance(other, Program):
W
Wu Yi 已提交
1754
            raise TypeError("_copy_param_info_from should be invoked with "
F
fengjiayi 已提交
1755 1756 1757
                            "Program")

        if len(self.blocks) != len(other.blocks):
W
Wu Yi 已提交
1758
            raise ValueError("_copy_param_info_from should be invoked with two "
F
fengjiayi 已提交
1759
                             "program, with represent the same topology")
1760
        for var in list(other.global_block().vars.values()):
F
fengjiayi 已提交
1761 1762 1763
            if var.is_data:
                self.global_block().var(var.name).is_data = True

1764
    def list_vars(self):
Y
yuyang18 已提交
1765 1766 1767 1768 1769 1770
        """
        Get all variables from this Program. A iterable object is returned.

        Returns:
            iterable: The generator will yield every variable in this program.
        """
1771
        for each_block in self.blocks:
1772
            for each_var in list(each_block.vars.values()):
1773 1774
                yield each_var

Y
Yu Yang 已提交
1775

Y
Yu Yang 已提交
1776
class Parameter(Variable):
1777
    """
1778
    Parameter is derived from Variable. A parameter is a persistable
1779
    Variable, and will be updated by optimizers after each iteration.
1780
    The training of a neural network is essentially the updating of
1781 1782
    its parameters.

1783
    Relative to a general Variable, a Parameter has several its own
1784 1785
    member variables:

1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797
    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.
1798 1799
    """

Y
Yu Yang 已提交
1800 1801 1802 1803 1804 1805 1806 1807 1808 1809
    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")
1810 1811 1812

        Variable.__init__(
            self, block, persistable=True, shape=shape, dtype=dtype, **kwargs)
Y
Yu Yang 已提交
1813 1814 1815 1816
        self.trainable = kwargs.get('trainable', True)

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

1817 1818
        self.regularizer = kwargs.get('regularizer', None)

F
fengjiayi 已提交
1819
        self.gradient_clip_attr = kwargs.get('gradient_clip_attr', None)
Y
Yu Yang 已提交
1820

W
wanghaoshuang 已提交
1821
        self.do_model_average = kwargs.get('do_model_average', None)
W
wanghaoshuang 已提交
1822

F
fengjiayi 已提交
1823 1824 1825
    def __str__(self):
        return self.to_string(True)

F
update  
fengjiayi 已提交
1826 1827 1828
    def to_string(self, throw_on_error, with_details=False):
        """
        To debug string.
D
dzhwinter 已提交
1829

F
update  
fengjiayi 已提交
1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843
        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
1844
                               "gradient_clip_attr", "do_model_average")
F
update  
fengjiayi 已提交
1845
            for attr_name in additional_attr:
1846 1847
                res_str += "%s: %s\n" % (
                    attr_name, six.binary_type(getattr(self, attr_name)))
F
update  
fengjiayi 已提交
1848 1849
        else:
            res_str = Variable.to_string(self, throw_on_error, False)
F
fengjiayi 已提交
1850 1851 1852 1853
        return res_str

    __repr__ = __str__

Y
Yu Yang 已提交
1854

Y
Yu Yang 已提交
1855
# program is a global instance.
Y
Yu Yang 已提交
1856 1857
_main_program_ = Program()
_startup_program_ = Program()
1858

1859

1860
def default_startup_program():
Y
Yu Yang 已提交
1861
    """
Y
yuyang18 已提交
1862 1863 1864 1865 1866 1867 1868 1869 1870
    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.
1871

Y
Yu Yang 已提交
1872 1873 1874
    Returns:
        Program: startup program
    """
Y
Yu Yang 已提交
1875
    return _startup_program_
1876

1877

1878
def default_main_program():
Y
Yu Yang 已提交
1879
    """
Y
yuyang18 已提交
1880 1881 1882 1883 1884 1885 1886 1887 1888
    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.
1889

Y
Yu Yang 已提交
1890 1891 1892
    Returns:
        Program: main program
    """
Y
Yu Yang 已提交
1893
    return _main_program_
Y
Yu Yang 已提交
1894 1895 1896 1897 1898


def switch_main_program(program):
    """
    Switch the main program to a new program.
1899

Y
Yu Yang 已提交
1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913
    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):
    """
1914
    Switch the startup program to a new program
Y
Yu Yang 已提交
1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929
    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
1930 1931 1932
    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.
1933

Y
Yu Yang 已提交
1934
    Examples:
Y
yuyang18 已提交
1935 1936 1937 1938 1939 1940 1941 1942 1943 1944

        >>> 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.
1945

Y
Yu Yang 已提交
1946
    Examples:
Y
yuyang18 已提交
1947 1948 1949 1950 1951 1952

        >>> 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 = ...
1953

Y
Yu Yang 已提交
1954
    Args:
Y
yuyang18 已提交
1955
        main_program(Program): New main program inside `with` statement.
1956
        startup_program(Program): New startup program inside `with` statement.
Y
Yu Yang 已提交
1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969
            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
1970 1971 1972 1973


def get_var(name, program=None):
    """
Y
yuyang18 已提交
1974
    Get a variable by name from the global block of a program.
F
fengjiayi 已提交
1975

X
xuwei06 已提交
1976 1977 1978
    Args:
        name(str): name of the variable
        program(Program|None): program object.
T
tangwei12 已提交
1979
        If None, default_global_program() will be used.
X
xuwei06 已提交
1980 1981 1982 1983 1984 1985 1986

    Returns:
        Variable
    """
    if program is None:
        program = default_main_program()
    assert isinstance(name, str)
1987
    assert isinstance(program, Program)
X
xuwei06 已提交
1988 1989

    return program.global_block().var(name)