framework.py 71.0 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.

15 16
from __future__ import print_function

Y
Yu Yang 已提交
17
import collections
Q
qiaolongfei 已提交
18
import contextlib
F
fengjiayi 已提交
19
import re
20
import six
21

Y
Yu Yang 已提交
22
import numpy as np
Q
qiaolongfei 已提交
23

M
minqiyang 已提交
24
from .. import compat as cpt
25
from .proto import framework_pb2
26 27
try:
    from . import core
28
except ImportError as e:
29 30 31 32
    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
M
minqiyang 已提交
33
    directory. The original error is: \n""" + cpt.get_exception_message(e))
34
except Exception as e:
35
    raise e
36
from . import unique_name
Y
Yu Yang 已提交
37

38
__all__ = [
39 40 41 42 43
    'Program',
    'Operator',
    'default_startup_program',
    'default_main_program',
    'program_guard',
X
xuwei06 已提交
44
    'get_var',
45
    'name_scope',
46
]
Y
Yu Yang 已提交
47

Q
qiaolongfei 已提交
48 49 50 51
EMPTY_VAR_NAME = core.kEmptyVarName()
TEMP_VAR_NAME = core.kTempVarName()
GRAD_VAR_SUFFIX = core.kGradVarSuffix()
ZERO_VAR_SUFFIX = core.kZeroVarSuffix()
W
Wu Yi 已提交
52 53 54
CONTROL_DEP_VAR_PREFIX = core.kControlDepVarName()


55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118
class NameScope(object):
    def __init__(self, name="", parent=None):
        self._children = dict()
        self._name = name
        self._parent = parent

    def child(self, prefix):
        if prefix not in self._children:
            new_child = NameScope(prefix, self)
            self._children[prefix] = [new_child]
        else:
            new_child = NameScope(prefix + "_%d" % len(self._children[prefix]),
                                  self)
            self._children[prefix].append(new_child)
        return new_child

    def parent(self):
        return self._parent

    def name(self):
        return self._name


_name_scope = NameScope()


@contextlib.contextmanager
def name_scope(prefix=None):
    """
    Generate hierarchical name prefix for the operators.

    Note: This should only used for debugging and visualization purpose.
    Don't use it for serious analysis such as graph/program transformations.

    Args:
        prefix(str): prefix.

    Examples:
        .. code-block:: python
          with name_scope("encoder"):
             ...
          with name_scope("decoder"):
             ...
             with name_scope("attention"):
                ...
    """
    # TODO(panyx0718): Only [0-9a-z].
    assert prefix, "namescope prefix cannot be empty."
    global _name_scope
    _name_scope = _name_scope.child(prefix)
    yield
    _name_scope = _name_scope.parent()


def _full_name_scope():
    global _name_scope
    scope = _name_scope
    name = ""
    while scope:
        name = scope.name() + "/" + name
        scope = scope.parent()
    return name


W
Wu Yi 已提交
119 120 121
def generate_control_dev_var_name():
    import random
    return CONTROL_DEP_VAR_PREFIX + "@" + str(random.random())
Q
qiaolongfei 已提交
122 123 124 125


def grad_var_name(var_name):
    """
126 127
    Returns:
        str: gradient name for a certain var name
Q
qiaolongfei 已提交
128 129 130
    """
    return var_name + GRAD_VAR_SUFFIX

Y
Yu Yang 已提交
131

132
def convert_np_dtype_to_dtype_(np_dtype):
133 134
    """
    Convert the data type in numpy to the data type in Paddle
135

136
    Args:
137
        np_dtype(np.dtype): the data type in numpy.
138

139 140
    Returns:
        core.VarDesc.VarType: the data type in Paddle.
141 142

    """
143 144
    dtype = np.dtype(np_dtype)
    if dtype == np.float32:
145
        return core.VarDesc.VarType.FP32
146
    elif dtype == np.float64:
147
        return core.VarDesc.VarType.FP64
148
    elif dtype == np.float16:
149
        return core.VarDesc.VarType.FP16
150
    elif dtype == np.int32:
151
        return core.VarDesc.VarType.INT32
152
    elif dtype == np.int16:
153
        return core.VarDesc.VarType.INT16
154
    elif dtype == np.int64:
155
        return core.VarDesc.VarType.INT64
156
    elif dtype == np.bool:
157
        return core.VarDesc.VarType.BOOL
158 159
    elif dtype == np.uint16:
        return core.VarDesc.VarType.INT16
160 161
    elif dtype == np.uint8:
        return core.VarDesc.VarType.UINT8
Q
qingqing01 已提交
162 163
    elif dtype == np.int8:
        return core.VarDesc.VarType.INT8
164
    else:
M
minqiyang 已提交
165
        raise ValueError("Not supported numpy dtype %s" % dtype)
166 167 168


def dtype_is_floating(dtype):
169 170 171
    """
    Check the data type is floating or not.
    Args:
172
        dtype(np.dtype|core.VarDesc.VarType): data type.
173 174 175 176 177
            Could be numpy format or Paddle format

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

    """
178
    if not isinstance(dtype, core.VarDesc.VarType):
179 180
        dtype = convert_np_dtype_to_dtype_(dtype)

181 182 183 184
    return dtype in [
        core.VarDesc.VarType.FP16, core.VarDesc.VarType.FP32,
        core.VarDesc.VarType.FP64
    ]
185 186


Y
Yang Yang(Tony) 已提交
187
def _debug_string_(proto, throw_on_error=True):
188 189 190 191 192 193 194 195 196 197 198
    """
    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
199
    error_fields = list()
Y
Yang Yang(Tony) 已提交
200
    if not proto.IsInitialized(error_fields) and throw_on_error:
C
caoying03 已提交
201 202
        raise ValueError("{0} are not initialized.\nThe message is {1}:\n".
                         format(error_fields, proto))
Y
Yu Yang 已提交
203 204 205
    return proto.__str__()


Y
Yu Yang 已提交
206
class Variable(object):
207
    """
208 209 210
    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
211
    two variables in different blocks could have the same name.
212

213 214
    There are many kinds of variables. Each kind of them has its own attributes
    and usages. Please reference the framework.proto for details.
215

216
    Most of a Variable's member variables can be setted to be None. It mean
217
    it is not available or will be specified later.
218 219

    Args:
220
        block(Block): The block that the variable belongs to.
221 222
        type(core.VarDesc.VarType): Variable type. Please reference the
            framework.proto for details.
223 224
        name(str|None): The name of the variable. If setted None, it will be
            generated automatically. Default: None
225
        shape(tuple|list|None): The shape of the variable. -1 means the batch size.
226
            Some kinds of variable do not contain shape, just set it to None.
227 228 229
            Default: None
        dtype(np.dtype|core.VarDesc.VarType|str|None): The data type of variable.
            Default: None
230
        lod_level (int|None): The level of lod tensor. 0 means it is not a time
231
            series data.
232
            Default: None
233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254
        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')
255 256
    """

Y
Yu Yang 已提交
257 258
    def __init__(self,
                 block,
Y
Yu Yang 已提交
259
                 type=core.VarDesc.VarType.LOD_TENSOR,
Y
Yu Yang 已提交
260 261 262 263
                 name=None,
                 shape=None,
                 dtype=None,
                 lod_level=None,
264
                 capacity=None,
Q
QI JUN 已提交
265
                 persistable=None,
F
fengjiayi 已提交
266
                 error_clip=None,
Y
Yu Yang 已提交
267
                 stop_gradient=False,
F
fengjiayi 已提交
268
                 is_data=False,
Y
Yu Yang 已提交
269
                 **kwargs):
Y
Yu Yang 已提交
270
        self.block = block
F
fengjiayi 已提交
271
        self.error_clip = error_clip
Y
Yu Yang 已提交
272 273

        if name is None:
Y
Yu Yang 已提交
274
            name = unique_name.generate('_generated_var')
D
Dong Zhihong 已提交
275
        is_new_var = False
M
minqiyang 已提交
276
        name = cpt.to_text(name)
M
minqiyang 已提交
277
        self.desc = self.block.desc.find_var(cpt.to_bytes(name))
D
Dong Zhihong 已提交
278 279

        if self.desc is None:
M
minqiyang 已提交
280
            self.desc = self.block.desc.var(cpt.to_bytes(name))
Y
Yu Yang 已提交
281
            is_new_var = True
Y
Yu Yang 已提交
282

Y
Yu Yang 已提交
283 284 285 286 287 288 289 290
        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
291
        if shape is not None:
Y
Yu Yang 已提交
292
            if is_new_var:
293
                self.desc.set_shape(shape)
Y
Yu Yang 已提交
294 295 296 297 298 299 300 301
            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
302
        if dtype is not None:
303
            if not isinstance(dtype, core.VarDesc.VarType):
304
                dtype = convert_np_dtype_to_dtype_(dtype)
Y
Yu Yang 已提交
305
            if is_new_var:
F
fengjiayi 已提交
306
                self.desc.set_dtype(dtype)
Y
Yu Yang 已提交
307
            else:
F
fengjiayi 已提交
308
                old_dtype = self.dtype
Q
QI JUN 已提交
309
                if dtype != old_dtype:
Y
Yu Yang 已提交
310 311 312 313 314
                    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
315 316

        if lod_level is not None:
Y
Yu Yang 已提交
317
            if is_new_var:
318
                self.desc.set_lod_level(lod_level)
Y
Yu Yang 已提交
319 320 321 322 323 324 325
            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))
326 327 328 329 330 331 332 333 334 335 336
        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))

337 338 339 340 341 342 343 344
        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
345
        self.block.vars[name] = self
Y
Yu Yang 已提交
346
        self.op = None
Y
Yu Yang 已提交
347
        self.stop_gradient = stop_gradient
F
fengjiayi 已提交
348
        self.is_data = is_data
Y
Yu Yang 已提交
349

350
    def __str__(self):
Y
Yang Yang(Tony) 已提交
351 352
        return self.to_string(True)

F
update  
fengjiayi 已提交
353
    def to_string(self, throw_on_error, with_details=False):
354 355 356 357
        """
        Get debug string.

        Args:
358 359
            throw_on_error(bool): True if raise an exception when self is
                not initialized.
F
update  
fengjiayi 已提交
360
            with_details(bool): more details about variables and parameters
361 362
                (e.g. trainable, optimize_attr, ...) will be printed when
                with_details is True. Default False;
363

364 365
        Returns:
            str: The debug string.
366
        """
F
update  
fengjiayi 已提交
367 368
        assert isinstance(throw_on_error, bool) and isinstance(with_details,
                                                               bool)
369
        protostr = self.desc.serialize_to_string()
370
        proto = framework_pb2.VarDesc.FromString(six.binary_type(protostr))
F
update  
fengjiayi 已提交
371 372 373 374
        res_str = _debug_string_(proto, throw_on_error)
        if with_details:
            additional_attr = ("error_clip", "stop_gradient")
            for attr_name in additional_attr:
375 376
                res_str += "%s: %s\n" % (
                    attr_name, six.binary_type(getattr(self, attr_name)))
F
update  
fengjiayi 已提交
377
        return res_str
378 379 380

    __repr__ = __str__

W
Wu Yi 已提交
381
    def _set_desc(self, input):
382 383 384 385 386 387 388 389 390
        """
        Set the variable description.

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

        Returns:
            None
        """
391 392
        self.desc = input

393 394 395 396
    @property
    def persistable(self):
        return self.desc.persistable()

Y
Yu Yang 已提交
397 398 399 400
    @persistable.setter
    def persistable(self, p):
        self.desc.set_persistable(p)

Y
Yu Yang 已提交
401 402
    @property
    def name(self):
M
minqiyang 已提交
403
        return cpt.to_text(self.desc.name())
Y
Yu Yang 已提交
404

T
typhoonzero 已提交
405 406 407 408
    @name.setter
    def name(self, new_name):
        self.desc.set_name(new_name)

Y
Yu Yang 已提交
409 410 411
    @property
    def shape(self):
        # convert to tuple, make it as same as numpy API.
412
        return tuple(self.desc.shape())
Y
Yu Yang 已提交
413 414

    @property
F
fengjiayi 已提交
415 416
    def dtype(self):
        return self.desc.dtype()
Y
Yu Yang 已提交
417 418 419

    @property
    def lod_level(self):
420
        return self.desc.lod_level()
Y
Yu Yang 已提交
421

Y
Yu Yang 已提交
422 423 424 425
    @property
    def type(self):
        return self.desc.type()

W
Wu Yi 已提交
426
    def _set_error_clip(self, error_clip):
427 428 429 430 431 432 433 434 435
        """
        Set the error_clip.

        Args:
            error_clip(BaseErrorClipAttr) : The new error_clip.

        Returns:
            None
        """
436 437
        self.error_clip = error_clip

Y
Yu Yang 已提交
438

F
fengjiayi 已提交
439 440 441
def get_all_op_protos():
    """
    Get all registered op proto from PaddlePaddle C++ end.
442

443 444
    Returns:
       list: list of OpProto.
F
fengjiayi 已提交
445 446 447 448
    """
    protostrs = core.get_all_op_protos()
    ret_values = []
    for pbstr in protostrs:
449
        op_proto = framework_pb2.OpProto.FromString(six.binary_type(pbstr))
F
fengjiayi 已提交
450 451 452 453 454
        ret_values.append(op_proto)
    return ret_values


class OpProtoHolder(object):
455 456 457 458
    """
    A global variable to hold all OpProtos from C++ as a map
    """

F
fengjiayi 已提交
459 460 461 462 463 464 465 466 467
    @classmethod
    def instance(cls):
        if not hasattr(cls, '_instance'):
            cls._instance = cls()
        return cls._instance

    def __init__(self):
        assert not hasattr(
            self.__class__,
468
            '_instance'), 'Please use `instance()` to get OpProtoHolder object!'
F
fengjiayi 已提交
469 470 471 472 473 474
        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):
475 476 477 478 479 480 481 482
        """
        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
483 484
        if type not in self.op_proto_map:
            raise ValueError("Operator \"%s\" has not been registered." % type)
F
fengjiayi 已提交
485 486
        return self.op_proto_map[type]

487 488 489 490
    @staticmethod
    def generated_op_attr_names():
        return {
            core.op_proto_and_checker_maker.kOpRoleAttrName(),
S
sneaxiy 已提交
491 492
            core.op_proto_and_checker_maker.kOpRoleVarAttrName(),
            core.op_proto_and_checker_maker.kOpNameScopeAttrName()
493 494
        }

F
fengjiayi 已提交
495

Y
Yu Yang 已提交
496
class Operator(object):
497
    """
498 499 500 501 502 503 504
    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
505
        type(str): The type of operator. Default None.
506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525
        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
526
        Block.append_op or Block._prepend_op instead.
527 528 529 530 531 532 533 534 535 536

    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]})
537
    """
538 539 540 541 542
    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
543
        'channel_recv', 'select', 'checkpoint_notify', 'gen_nccl_id'
544
    }
545

Y
Yu Yang 已提交
546 547
    def __init__(self,
                 block,
Y
Yu Yang 已提交
548
                 desc,
Y
Yu Yang 已提交
549 550 551 552 553
                 type=None,
                 inputs=None,
                 outputs=None,
                 attrs=None):
        self.block = block
Y
Yu Yang 已提交
554
        self.desc = desc
G
gongweibao 已提交
555 556 557 558 559
        # note: not add self.attrs here:
        # https://github.com/PaddlePaddle/Paddle/pull/12583#pullrequestreview-145093173
        op_attrs = attrs
        if op_attrs is None:
            op_attrs = dict()
Y
yuyang18 已提交
560 561 562 563
        del attrs

        op_maker = core.op_proto_and_checker_maker

G
gongweibao 已提交
564 565
        if op_maker.kOpRoleAttrName() not in op_attrs:
            op_attrs[op_maker.kOpRoleAttrName()] = self.block.program.op_role
Y
yuyang18 已提交
566 567 568

        role_var_name = op_maker.kOpRoleVarAttrName()
        if len(self.block.program.
G
gongweibao 已提交
569 570
               op_role_var) != 0 and role_var_name not in op_attrs:
            op_attrs[role_var_name] = self.block.program.op_role_var
Y
yuyang18 已提交
571

G
gongweibao 已提交
572 573
        if role_var_name in op_attrs and len(op_attrs[role_var_name]) == 0:
            del op_attrs[role_var_name]
Y
yuyang18 已提交
574

F
fengjiayi 已提交
575 576 577 578 579
        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
580
        self.desc.set_type(type)
F
fengjiayi 已提交
581
        proto = OpProtoHolder.instance().get_op_proto(type)
582

583 584 585
        namescope_var_name = op_maker.kOpNameScopeAttrName()
        op_attrs[namescope_var_name] = _full_name_scope()

Y
Yang Yang(Tony) 已提交
586 587
        def find_name(var_list, name):
            for var_name in var_list:
Q
Qiao Longfei 已提交
588
                if var_list[var_name] is not None and var_name == name:
Y
Yang Yang(Tony) 已提交
589 590
                    return True
            return False
Q
QI JUN 已提交
591

Y
Yang Yang(Tony) 已提交
592 593 594 595 596 597 598
        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:
599 600 601 602
                    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
603 604
                        raise ValueError(
                            "Input %s expects only one input, but %d are given."
605 606 607
                            % (in_proto.name, len(in_args)))
                    in_arg_names = []
                    for arg in in_args:
608
                        if isinstance(arg, six.string_types):
Y
Yang Yu 已提交
609
                            in_arg_names.append(arg)
610 611
                        elif isinstance(arg, six.binary_type):
                            in_arg_names.append(arg.decode())
Y
Yang Yu 已提交
612
                        else:
M
minqiyang 已提交
613
                            in_arg_names.append(cpt.to_text(arg.name))
614
                    self.desc.set_input(in_proto.name, in_arg_names)
Y
Yang Yang(Tony) 已提交
615 616
                else:
                    self.desc.set_input(in_proto.name, [])
F
Update  
fengjiayi 已提交
617

Y
Yu Yang 已提交
618
        if outputs is not None:
619 620 621 622 623 624 625
            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
626 627
                raise ValueError(("Incorrect setting for output(s) of "
                                  "operator \"%s\". Need: [%s] Given: [%s]") %
628 629 630
                                 (type,
                                  ", ".join(six.binary_type(e) for e in need),
                                  ", ".join(six.binary_type(e) for e in given)))
631

F
fengjiayi 已提交
632
            for out_proto in proto.outputs:
633 634 635 636
                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
637 638
                    raise ValueError(
                        "Output %s expects only one output, but %d are given." %
639 640 641
                        (out_proto.name, len(out_args)))
                out_arg_names = []
                for arg in out_args:
M
minqiyang 已提交
642
                    out_arg_names.append(cpt.to_text(arg.name))
643 644
                    arg.op = self
                self.desc.set_output(out_proto.name, out_arg_names)
F
Update  
fengjiayi 已提交
645

G
gongweibao 已提交
646 647
        if op_attrs is not None:
            if not isinstance(op_attrs, dict):
648
                raise TypeError("'attrs' should be a dict.")
F
fengjiayi 已提交
649
            for attr in proto.attrs:
F
Update  
fengjiayi 已提交
650
                attr_name = attr.name
G
gongweibao 已提交
651
                if (attr_name not in op_attrs) or (op_attrs[attr_name] is None):
F
Update  
fengjiayi 已提交
652
                    continue
G
gongweibao 已提交
653
                attr_val = op_attrs[attr_name]
G
gongweibao 已提交
654 655
                self._update_desc_attr(attr_name, attr_val)

656
        self.desc.check_attrs()
657
        if self.has_kernel(type):
Q
QI JUN 已提交
658
            self.desc.infer_var_type(self.block.desc)
Y
Yu Yang 已提交
659
            self.desc.infer_shape(self.block.desc)
F
fengjiayi 已提交
660

661 662 663
    def has_kernel(self, op_type):
        return op_type not in self.OP_WITHOUT_KERNEL_SET

Y
Yang Yang(Tony) 已提交
664
    def to_string(self, throw_on_error):
665
        """
666 667
        Get debug string.

668
        Args:
669 670
            throw_on_error(bool): Whether to raise exception if self is not
                initialized.
671

672 673
        Returns:
            str: The debug string.
674 675

        """
676
        protostr = self.desc.serialize_to_string()
677
        proto = framework_pb2.OpDesc.FromString(six.binary_type(protostr))
Y
Yang Yang(Tony) 已提交
678 679 680 681
        return _debug_string_(proto, throw_on_error)

    def __str__(self):
        return self.to_string(True)
682 683 684

    __repr__ = __str__

F
fengjiayi 已提交
685 686 687 688 689
    @property
    def type(self):
        return self.desc.type()

    def input(self, name):
690
        """
691
        Get the input arguments according to the input parameter name.
692

693 694
        Args:
            name(str): The input parameter name.
695

696 697 698
        Returns:
            list: return the list of argument names that associated with \
                the specific parameter name.
699
        """
F
fengjiayi 已提交
700 701
        return self.desc.input(name)

T
typhoonzero 已提交
702
    def rename_input(self, old_name, new_name):
703 704 705 706 707 708 709 710 711 712
        """
        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
713 714 715
        self.desc.rename_input(old_name, new_name)

    def rename_output(self, old_name, new_name):
716 717 718 719 720 721 722 723 724 725
        """
        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
726 727
        self.desc.rename_output(old_name, new_name)

F
fengjiayi 已提交
728 729 730 731
    @property
    def input_names(self):
        return self.desc.input_names()

T
typhoonzero 已提交
732 733 734 735 736 737 738 739
    @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
740
    def output(self, name):
741
        """
742
        Get output arguments by the output parameter name.
743

744 745
        Args:
            name(str): The output parameter name.
746

747 748 749
        Returns:
            list: return the list of argument names associated with \
                the specific parameter name.
750
        """
F
fengjiayi 已提交
751 752 753 754 755 756
        return self.desc.output(name)

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

757 758 759 760 761 762 763 764
    @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
765
    def has_attr(self, name):
766
        """
767 768
        Whether this Operator has the attribute with name or not.

769
        Args:
770
            name(str): the attribute name.
771

772 773
        Returns:
            bool: True if has this attribute.
774 775

        """
F
fengjiayi 已提交
776 777 778
        return self.desc.has_attr(name)

    def attr_type(self, name):
779
        """
780
        Get the type of attribute by attribute's name.
781

782 783
        Args:
            name(str): the attribute name.
784

785 786
        Returns:
            core.AttrType: the attribute type.
787
        """
F
fengjiayi 已提交
788 789
        return self.desc.attr_type(name)

Y
yuyang18 已提交
790
    def set_attr(self, name, val):
791 792 793 794 795 796 797 798 799 800
        """
        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).
        """
G
gongweibao 已提交
801 802 803 804 805 806 807 808 809 810 811 812 813
        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
814 815
        if isinstance(val, Block):
            self.desc.set_block_attr(name, val.desc)
Y
Yancey1989 已提交
816 817
        elif isinstance(val, list) and val and all(
                isinstance(v, Block) for v in val):
818
            self.desc.set_blocks_attr(name, [v.desc for v in val])
Q
Qiyang Min 已提交
819 820 821 822 823
        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
824

F
fengjiayi 已提交
825 826 827 828 829
    @property
    def attr_names(self):
        return self.desc.attr_names()

    def attr(self, name):
830
        """
831 832
        Get the attribute by name.

833
        Args:
834
            name(str): the attribute name.
835

836 837
        Returns:
            bool|int|str|float|list: The attribute value. The return value
838 839
            can be any valid attribute type.
        """
F
fengjiayi 已提交
840
        return self.desc.attr(name)
Y
Yu Yang 已提交
841

G
gongweibao 已提交
842
    def block_attr_id(self, name):
843
        """
G
gongweibao 已提交
844
        Get the block attribute's id by name.
845

846 847
        Args:
            name(str): the attribute name.
848

849 850
        Returns:
            int: the block index.
851
        """
G
gongweibao 已提交
852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897
        return self.desc.block_attr_id(name)

    def block_attr(self, name):
        """
        Get the block attribute  by name.

        Args:
            name(str): the attribute name.

        Returns:
            block: the block attribute.
        """

        id = self.block_attr_id(name)
        assert (id >= 0 and id < len(self.block.program.blocks))
        return self.block.program.blocks[id]

    def blocks_attr(self, name):
        """
        Get the blocks attribute  by name.

        Args:
            name(str): the attribute name.

        Returns:
            list: list of the blocks attribute.
        """
        attrs = []
        for i in self.blocks_attr_ids(name):
            assert (i >= 0 and i < len(self.block.program.blocks))
            attrs.append(self.block.program.blocks[i])

        return attrs

    def blocks_attr_ids(self, name):
        """
        Get the blocks attribute's ids by name.

        Args:
            name(str): the attribute name.

        Returns:
            list: list of the blocks ids.
        """

        return self.desc.blocks_attr_ids(name)
Y
Yu Yang 已提交
898

J
JiayiFeng 已提交
899
    def all_attrs(self):
F
fengjiayi 已提交
900
        """
901 902 903
        Get the attribute dict.

        Returns:
G
gongweibao 已提交
904
            dict: The Operator's attribute dict, name->attr.
F
fengjiayi 已提交
905 906 907 908
        """
        attr_names = self.attr_names
        attr_map = {}
        for n in attr_names:
G
gongweibao 已提交
909 910
            attr_type = self.desc.attr_type(n)
            if attr_type == core.AttrType.BLOCK:
F
fengjiayi 已提交
911
                attr_map[n] = self.block_attr(n)
G
gongweibao 已提交
912 913 914 915 916 917 918 919
                continue

            if attr_type == core.AttrType.BLOCKS:
                attr_map[n] = self.blocks_attr(n)
                continue

            attr_map[n] = self.attr(n)

F
fengjiayi 已提交
920 921
        return attr_map

Y
Yu Yang 已提交
922

Y
Yu Yang 已提交
923
class Block(object):
924 925 926 927 928 929 930 931 932 933 934 935 936 937
    """
    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
W
Wu Yi 已提交
938
        use `Program._create_block()` to create a block.
939 940 941 942 943 944 945 946 947 948 949 950 951 952

    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
953
    def __init__(self, program, idx):
Y
Yu Yang 已提交
954
        self.desc = program.desc.block(idx)
955
        self.vars = collections.OrderedDict()  # var_name --> var
Q
qiaolongfei 已提交
956
        self.ops = list()  # operator list
Y
Yu Yang 已提交
957
        self.program = program
958
        self.removed_vars = collections.OrderedDict()
Y
Yu Yang 已提交
959

960
    def __str__(self):
Y
Yang Yang(Tony) 已提交
961 962
        return self.to_string(True)

F
fengjiayi 已提交
963 964
    def to_string(self, throw_on_error, with_details=False):
        """
965 966
        Get debug string.

F
fengjiayi 已提交
967 968
        Args:
            throw_on_error(bool): raise exception when self is not initialized
969
                when throw_on_error is True.
F
update  
fengjiayi 已提交
970
            with_details(bool): more details about variables and parameters
971 972
                (e.g. trainable, optimize_attr, ...) will be printed when
                with_details is True. Default False.
F
fengjiayi 已提交
973

974 975
        Returns:
            str: The debug string.
F
fengjiayi 已提交
976 977 978 979
        """
        assert isinstance(throw_on_error, bool) and isinstance(with_details,
                                                               bool)
        if with_details:
F
fengjiayi 已提交
980
            re_add_indent = re.compile(r"\n(.)")
F
fengjiayi 已提交
981 982
            res_str = "blocks {\n  idx: %d\n  parent_idx: %d" % (
                self.idx, self.parent_idx)
983
            for var in list(self.vars.values()):
F
fengjiayi 已提交
984
                res_str += "\n  vars {\n    %s  }" % re_add_indent.sub(
F
update  
fengjiayi 已提交
985
                    r"\n    \1", var.to_string(throw_on_error, with_details))
F
fengjiayi 已提交
986
            for op in self.ops:
F
fengjiayi 已提交
987 988
                res_str += "\n  ops {\n    %s  }" % re_add_indent.sub(
                    r"\n    \1", op.to_string(throw_on_error))
F
fengjiayi 已提交
989 990 991
            res_str += "\n}"
        else:
            protostr = self.desc.serialize_to_string()
992 993
            proto = framework_pb2.BlockDesc.FromString(
                six.binary_type(protostr))
F
fengjiayi 已提交
994 995
            res_str = _debug_string_(proto, throw_on_error)
        return res_str
996 997 998

    __repr__ = __str__

Y
Yu Yang 已提交
999 1000
    @property
    def parent_idx(self):
Y
Yu Yang 已提交
1001
        return self.desc.parent
Y
Yu Yang 已提交
1002

Y
Yu Yang 已提交
1003 1004 1005 1006
    @property
    def forward_block_idx(self):
        return self.desc.get_forward_block_idx()

W
Wu Yi 已提交
1007
    def _set_forward_block_idx(self, idx):
1008 1009 1010 1011 1012 1013 1014 1015 1016
        """
        Set the forward block Idx.

        Args:
            idx(int): the block index.

        Returns:
            None
        """
W
Wu Yi 已提交
1017
        self.desc._set_forward_block_idx(idx)
Y
Yu Yang 已提交
1018

Y
Yu Yang 已提交
1019 1020
    @property
    def idx(self):
Y
Yu Yang 已提交
1021
        return self.desc.id
Y
Yu Yang 已提交
1022

Q
Qiao Longfei 已提交
1023
    def var(self, name):
1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036
        """
        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.
        """
1037
        if not isinstance(name, six.string_types):
M
minqiyang 已提交
1038 1039 1040
            raise TypeError(
                "var require string as parameter, but get %s instead." %
                (type(name)))
Y
Yu Yang 已提交
1041 1042
        v = self.vars.get(name, None)
        if v is None:
Q
Qiao Longfei 已提交
1043
            raise ValueError("var %s not in this block" % name)
Y
Yu Yang 已提交
1044
        return v
Q
Qiao Longfei 已提交
1045

W
Wu Yi 已提交
1046
    def _var_recursive(self, name):
1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059
        """
        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
1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085
        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
1086

Q
Qiao Longfei 已提交
1087
    def all_parameters(self):
1088
        return list(self.iter_parameters())
1089

1090
    def iter_parameters(self):
M
minqiyang 已提交
1091
        return (item[1] for item in six.iteritems(self.vars)
1092
                if isinstance(item[1], Parameter))
Q
Qiao Longfei 已提交
1093

Y
Yu Yang 已提交
1094
    def create_var(self, *args, **kwargs):
1095
        var = Variable(block=self, *args, **kwargs)
1096 1097
        if 'initializer' in kwargs:
            kwargs['initializer'](var, self)
Q
Qiao Longfei 已提交
1098
        return var
Y
Yu Yang 已提交
1099

Q
Qiao Longfei 已提交
1100 1101 1102
    def has_var(self, name):
        return name in self.vars

W
Wu Yi 已提交
1103
    def _rename_var(self, name, new_name):
T
typhoonzero 已提交
1104 1105
        """
        Rename variable in vars and ops' inputs and outputs
1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117

        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
1118
        """
M
minqiyang 已提交
1119 1120
        name = cpt.to_text(name)
        new_name = cpt.to_text(new_name)
M
minqiyang 已提交
1121

T
typhoonzero 已提交
1122
        if not self.has_var(name):
1123
            raise ValueError("var %s is not in current block" % name)
T
wip  
typhoonzero 已提交
1124 1125
        v = self.var(name)
        if type(v) == Parameter:
T
typhoonzero 已提交
1126
            var_type = "Parameter"
T
wip  
typhoonzero 已提交
1127 1128 1129 1130 1131 1132 1133
            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
1134
            var_type = "Variable"
T
wip  
typhoonzero 已提交
1135 1136 1137 1138
            error_clip = v.error_clip
            stop_gradient = v.stop_gradient
        else:
            raise ValueError("unsupported var type: %s", type(v))
T
typhoonzero 已提交
1139
        orig_var_type = v.type
M
minqiyang 已提交
1140
        self.desc._rename_var(cpt.to_bytes(name), cpt.to_bytes(new_name))
W
Wu Yi 已提交
1141
        # NOTE: v is destroyed by C++ after calling _rename_var.
M
minqiyang 已提交
1142
        d = self.desc.find_var(cpt.to_bytes(new_name))
T
typhoonzero 已提交
1143
        if var_type == "Parameter":
T
wip  
typhoonzero 已提交
1144 1145 1146 1147
            var = Parameter(
                self,
                d.shape(),
                d.dtype(),
T
typhoonzero 已提交
1148
                type=orig_var_type,
T
wip  
typhoonzero 已提交
1149 1150 1151 1152 1153 1154 1155
                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
1156
        elif var_type == "Variable":
T
wip  
typhoonzero 已提交
1157 1158
            var = Variable(
                self,
T
typhoonzero 已提交
1159
                type=orig_var_type,
T
wip  
typhoonzero 已提交
1160 1161 1162 1163
                name=new_name,
                error_clip=error_clip,
                stop_gradient=stop_gradient)

W
Wu Yi 已提交
1164
        # rename the python side, _sync_with_cpp will only add
T
wip  
typhoonzero 已提交
1165 1166 1167
        # new vars/ops to python side.
        self.vars[new_name] = var
        del self.vars[name]
W
Wu Yi 已提交
1168
        self._sync_with_cpp()
1169
        return var
T
typhoonzero 已提交
1170

W
Wu Yi 已提交
1171 1172
    def _remove_var(self, name):
        self._sync_with_cpp()
M
minqiyang 已提交
1173
        self.desc._remove_var(cpt.to_bytes(name))
1174 1175
        del self.vars[name]

Y
Yu Yang 已提交
1176 1177
    def create_parameter(self, *args, **kwargs):
        global_block = self.program.global_block()
Q
Qiao Longfei 已提交
1178
        param = Parameter(global_block, *args, **kwargs)
1179
        if 'initializer' in kwargs:
1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199

            def _is_inited_by(block, var):
                init_ops = []
                for op in block.ops:
                    if var.name in op.output_arg_names:
                        init_ops.append(op)
                return init_ops

            initializer = kwargs['initializer']
            init_ops = _is_inited_by(global_block, param)
            init_ops_len = len(init_ops)
            if init_ops_len > 1:
                raise RuntimeError("param " + param.name +
                                   " is inited by multiple init ops " + str(
                                       init_ops))
            elif init_ops_len == 1:
                #TODO already inited, do nothing, should log a warning
                pass
            else:
                initializer(param, self)
Q
Qiao Longfei 已提交
1200
        return param
Y
Yu Yang 已提交
1201

Y
Yu Yang 已提交
1202
    def append_op(self, *args, **kwargs):
1203 1204 1205 1206 1207 1208
        """
        Appends a new Operator according to the giving arguments.

        Returns:
            Operator: the append Operator.
        """
Y
Yu Yang 已提交
1209
        op_desc = self.desc.append_op()
1210
        op = Operator(block=self, desc=op_desc, *args, **kwargs)
Y
Yu Yang 已提交
1211 1212 1213
        self.ops.append(op)
        return op

W
Wu Yi 已提交
1214
    def _insert_op(self, index, *args, **kwargs):
1215 1216 1217 1218 1219 1220 1221 1222 1223
        """
        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
1224 1225
        self._sync_with_cpp()
        op_desc = self.desc._insert_op(index)
Q
qiaolongfei 已提交
1226 1227 1228 1229
        op = Operator(block=self, desc=op_desc, *args, **kwargs)
        self.ops.insert(index, op)
        return op

W
Wu Yi 已提交
1230
    def _remove_op(self, index):
1231 1232 1233 1234 1235 1236 1237 1238 1239
        """
        Remove the specific position operator.

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

        Returns:
            None
        """
W
Wu Yi 已提交
1240 1241
        self._sync_with_cpp()
        self.desc._remove_op(index, index + 1)
1242 1243
        del self.ops[index]

W
Wu Yi 已提交
1244
    def _slice_ops(self, start, end):
1245 1246 1247 1248 1249 1250 1251 1252 1253 1254
        """
        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
1255
        return self.ops[start:end]
Y
Yancey1989 已提交
1256

W
Wu Yi 已提交
1257 1258
    def _prepend_op(self, *args, **kwargs):
        op_desc = self.desc._prepend_op()
Y
Yu Yang 已提交
1259
        op = Operator(self, op_desc, *args, **kwargs)
Q
qiaolongfei 已提交
1260
        self.ops.insert(0, op)
Y
Yu Yang 已提交
1261 1262
        return op

W
Wu Yi 已提交
1263
    def _sync_with_cpp(self):
1264
        """
1265 1266
        Sync from the desc on the c++ end. This method is used to synchronize
        the c++ desc instance generated by backward.
1267
        """
Q
Qiao Longfei 已提交
1268 1269 1270 1271 1272
        # 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())

1273
        # sync variables removed from c++ end
1274
        for var in list(self.vars.keys()):
M
minqiyang 已提交
1275
            if not self.desc.find_var(cpt.to_bytes(var)):
1276 1277
                self.vars.pop(var)

Q
Qiao Longfei 已提交
1278
        # sync operators from cpp
1279 1280 1281 1282
        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
1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298
        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
1299 1300 1301 1302 1303

        # 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
1304
            self.ops.insert(0, op)
Q
Qiao Longfei 已提交
1305 1306 1307 1308 1309 1310 1311

        # 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)

1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324
        # 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
1325 1326 1327 1328
        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
1329
    def _copy_param_info_from(self, other):
1330
        """
1331 1332
        Copy the information of parameters from the other block.

1333
        Args:
1334 1335 1336 1337 1338
            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.
1339 1340 1341 1342 1343

        Returns:
            None
        """
        if not isinstance(other, Block):
W
Wu Yi 已提交
1344 1345
            raise TypeError(
                "_copy_param_info_from should be invoked with Block")
1346
        for p in other.iter_parameters():
1347 1348 1349
            assert isinstance(p, Parameter)
            v = self.vars.get(p.name, None)
            if v is None:
W
Wu Yi 已提交
1350
                raise ValueError("_copy_param_info_from should be invoked with "
1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362
                                 "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
1363
                gradient_clip_attr=p.gradient_clip_attr,
F
fengjiayi 已提交
1364
                error_clip=p.error_clip,
1365 1366 1367
                name=v.name)
            self.vars[new_p.name] = new_p

W
Wu Yi 已提交
1368
    def _clone_variable(self, var):
1369 1370
        """
        Clone a variable into current block.
1371

1372 1373 1374 1375
        Args:
            var: the variable to be cloned.

        Returns:
1376
            Variable: the new  variable cloned from 'var' in current block.
1377 1378
        """
        assert isinstance(var, Variable)
T
update  
typhoonzero 已提交
1379 1380 1381 1382 1383
        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
1384 1385
        elif var.type == core.VarDesc.VarType.RAW:
            ret_var = self.create_var(
T
tangwei12 已提交
1386
                name=var.name, persistable=var.persistable, type=var.type)
T
typhoonzero 已提交
1387 1388 1389 1390 1391 1392
        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
1393 1394
                persistable=True,
                is_data=var.is_data)
T
update  
typhoonzero 已提交
1395 1396 1397 1398 1399 1400 1401
        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
1402 1403
                persistable=True,
                is_data=var.is_data)
T
update  
typhoonzero 已提交
1404
        return ret_var
1405

Y
Yu Yang 已提交
1406 1407

class Program(object):
D
dzhwinter 已提交
1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418
    """
    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
1419
    default_main_program run in every mini batch and adjust the weights.
D
dzhwinter 已提交
1420 1421

    Returns:
Y
yuyang18 已提交
1422
        A empty program.
D
dzhwinter 已提交
1423 1424

    Examples:
Y
yuyang18 已提交
1425 1426 1427 1428 1429 1430
        >>> 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
1431 1432 1433

    """

1434 1435
    def __init__(self):
        self.desc = core.ProgramDesc()
Y
Yu Yang 已提交
1436 1437
        self.blocks = [Block(self, 0)]
        self.current_block_idx = 0
D
dzhwinter 已提交
1438
        self._seed = 0
Y
yuyang18 已提交
1439
        self._current_role = core.op_proto_and_checker_maker.OpRole.Forward
Y
yuyang18 已提交
1440
        self._op_role_var = []
T
tangwei12 已提交
1441 1442 1443 1444

        # for distribute
        self._is_distributed = False
        self._is_chief = False
T
tangwei12 已提交
1445
        self._slice_vars_and_attrs = []
T
tangwei12 已提交
1446 1447
        self._endpoints = []
        self._distributed_lookup_table = None
Y
yuyang18 已提交
1448 1449 1450

    @property
    def op_role(self):
Y
yuyang18 已提交
1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463
        """
        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
1464 1465 1466 1467 1468 1469 1470 1471
        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
1472 1473 1474 1475 1476 1477 1478
        """
        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
1479 1480 1481 1482
        return self._op_role_var

    @op_role_var.setter
    def set_op_role_var(self, var_name):
Y
yuyang18 已提交
1483
        self._op_role_var = [var_name]
Y
yuyang18 已提交
1484 1485

    @contextlib.contextmanager
W
Wu Yi 已提交
1486
    def _optimized_guard(self, param_and_grads):
Y
yuyang18 已提交
1487 1488 1489 1490 1491 1492 1493
        """
        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:
1494
            param_and_grads(list): The variables (names) to be optimized.
Y
yuyang18 已提交
1495 1496 1497 1498

        Examples:

            >>> p, g = backward(...)
W
Wu Yi 已提交
1499
            >>> with program._optimized_guard([p,g]):
Y
yuyang18 已提交
1500 1501
            >>>     p = p - 0.001 * g
        """
Y
yuyang18 已提交
1502 1503
        OpRole = core.op_proto_and_checker_maker.OpRole
        self._current_role = OpRole.Optimize
1504 1505 1506 1507
        self._op_role_var = [
            var.name if isinstance(var, Variable) else var
            for var in param_and_grads
        ]
Y
yuyang18 已提交
1508
        yield
Y
yuyang18 已提交
1509
        self._op_role_var = []
Y
yuyang18 已提交
1510
        self._current_role = OpRole.Forward
Y
Yu Yang 已提交
1511

1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535
    @contextlib.contextmanager
    def _lr_schedule_guard(self):
        """
        A with guard to set :code:`LRSched` :code:`OpRole` and
        :code:`OpRoleVar` automatically. The :code:`OpRoleVar` is
        set to the target learning rate.

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


        Examples:

            >>> p, g = backward(...)
            >>> with program.lr_schedule_guard():
            >>>     lr = lr * decay
        """
        OpRole = core.op_proto_and_checker_maker.OpRole
        self._current_role = OpRole.LRSched
        # TODO(typhoonzero): how to set target learning rate var
        self._op_role_var = []
        yield
        self._op_role_var = []
        self._current_role = OpRole.Forward

1536
    def __str__(self):
Y
yuyang18 已提交
1537 1538 1539 1540 1541 1542 1543 1544 1545
        """
        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
1546 1547
        return self.to_string(True)

F
fengjiayi 已提交
1548 1549 1550
    def to_string(self, throw_on_error, with_details=False):
        """
        To debug string.
Y
yuyang18 已提交
1551

F
fengjiayi 已提交
1552
        Args:
Y
yuyang18 已提交
1553 1554
            throw_on_error(bool): raise Value error when any of required fields
                is not set.
F
fengjiayi 已提交
1555

Y
yuyang18 已提交
1556 1557 1558 1559 1560 1561 1562 1563 1564 1565
            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
1566 1567 1568 1569 1570 1571 1572 1573 1574 1575

        """
        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()
1576 1577
            proto = framework_pb2.ProgramDesc.FromString(
                six.binary_type(protostr))
F
fengjiayi 已提交
1578 1579
            res_str = _debug_string_(proto, throw_on_error)
        return res_str
1580

W
Wu Yi 已提交
1581
    def _get_desc(self):
Y
yuyang18 已提交
1582 1583 1584 1585 1586 1587 1588
        """
        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.
        """
1589 1590
        return self.desc

X
version  
Xin Pan 已提交
1591 1592 1593
    def _version(self):
        return self.desc._version()

1594
    def clone(self, for_test=False):
Y
yuyang18 已提交
1595 1596 1597
        """
        Create a new, duplicated program.

1598

Y
yuyang18 已提交
1599 1600 1601 1602
        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`.
1603

Y
yuyang18 已提交
1604 1605 1606 1607
        * 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
1608 1609 1610 1611 1612
        :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()
1613 1614

        Args:
Y
yuyang18 已提交
1615 1616
            for_test(bool): True if change the :code:`is_test` attribute of
                operators to :code:`True`.
1617

D
dzhwinter 已提交
1618
        Returns:
Y
yuyang18 已提交
1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671
            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.
1672 1673
        """
        if for_test:
X
Xin Pan 已提交
1674
            p = self._inference_optimize(prune_read_op=False)
1675
        else:
1676
            p = Program()
G
gongweibao 已提交
1677 1678
            p.current_block_idx = self.current_block_idx
            p._seed = self._seed
1679
            p.desc = core.ProgramDesc(self.desc)
M
minqiyang 已提交
1680 1681 1682
            p.blocks = [
                Block(p, i) for i in six.moves.range(self.desc.num_blocks())
            ]
G
gongweibao 已提交
1683 1684 1685 1686

            p._current_role = self._current_role
            p._op_role_var = self._op_role_var

W
Wu Yi 已提交
1687
            p._sync_with_cpp()
1688

W
Wu Yi 已提交
1689
        p._copy_param_info_from(self)
W
Wu Yi 已提交
1690
        p._copy_data_info_from(self)
Y
Yu Yang 已提交
1691
        return p
1692

W
Wu Yi 已提交
1693
    def _prune(self, targets):
Y
yuyang18 已提交
1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708
        """
        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.

        """
1709 1710 1711 1712 1713 1714
        if not isinstance(targets, list):
            targets = [targets]
        targets_idx = []
        for t in targets:
            if not isinstance(t, Operator):
                if isinstance(t, Variable):
1715 1716
                    # After transpiler processing, the op that output this
                    # variable maybe has been changed, so t.op is not reliable
1717
                    # and we need to find the current op that generate this
1718 1719 1720 1721 1722 1723 1724 1725
                    # 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

1726
                    t = t.op
1727 1728 1729 1730
                    if t is None:
                        raise ValueError(
                            "The target variable must have an "
                            "associated operator that generates it.")
1731
                else:
1732 1733
                    raise ValueError("All targets of prune() can only be "
                                     "Variable or Operator.")
1734 1735 1736 1737

            targets_idx.append([t.block.idx, t.idx])
        res = Program()
        res.desc = core.prune(self.desc, targets_idx)
M
minqiyang 已提交
1738 1739 1740
        res.blocks = [
            Block(res, i) for i in six.moves.range(res.desc.num_blocks())
        ]
W
Wu Yi 已提交
1741
        res._sync_with_cpp()
1742 1743
        return res

X
Xin Pan 已提交
1744
    def _inference_optimize(self, prune_read_op=True):
Y
yuyang18 已提交
1745
        """
F
fengjiayi 已提交
1746 1747 1748 1749 1750
        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.

1751
        3. change the :code:`is_test`
Y
yuyang18 已提交
1752 1753 1754
        attribute of operators to :code:`True`. All the :code:`Parameter`
        information will be lost.

1755
        Args:
X
Xin Pan 已提交
1756 1757
            prune_read_op(bool): remove the read ops that are added by py_reader
                                 for cpp inference library
1758

Y
yuyang18 已提交
1759 1760 1761 1762 1763 1764
        Notes: This API is a very low level API. Use
        :code:`Program.clone(for_test=True)` instead.

        Returns:
            Program: The new program.
        """
1765
        res = Program()
1766
        res.desc = core.ProgramDesc(self.desc)
F
fengjiayi 已提交
1767 1768 1769 1770

        # remove all readers and the read_op if exist
        read_op_idx = 0
        root_block = res.desc.block(0)
X
Xin Pan 已提交
1771
        if prune_read_op:
1772 1773 1774 1775 1776 1777 1778 1779 1780
            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:
M
minqiyang 已提交
1781
                    root_block._remove_var(cpt.to_bytes(var.name()))
F
fengjiayi 已提交
1782 1783

        # change all `is_test` attributes to True
M
minqiyang 已提交
1784
        for i in six.moves.range(res.desc.num_blocks()):
1785
            block = res.desc.block(i)
M
minqiyang 已提交
1786
            for j in six.moves.range(block.op_size()):
1787 1788 1789
                op = block.op(j)
                if op.has_attr('is_test'):
                    op.set_attr('is_test', True)
M
minqiyang 已提交
1790 1791 1792
        res.blocks = [
            Block(res, i) for i in six.moves.range(res.desc.num_blocks())
        ]
W
Wu Yi 已提交
1793
        res._sync_with_cpp()
1794 1795
        return res

1796 1797
    @staticmethod
    def parse_from_string(binary_str):
Y
yuyang18 已提交
1798 1799 1800 1801 1802 1803 1804
        """
        Deserialize a program desc from protobuf binary string.

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

        Args:
1805
            binary_str_type(str): The binary prootbuf string.
Y
yuyang18 已提交
1806 1807 1808 1809

        Returns:
            Program: A deserialized program desc.
        """
1810 1811
        p = Program()
        p.desc = core.ProgramDesc(binary_str)
M
minqiyang 已提交
1812
        p.blocks = [Block(p, i) for i in six.moves.range(p.desc.num_blocks())]
W
Wu Yi 已提交
1813
        p._sync_with_cpp()
1814
        return p
Y
Yu Yang 已提交
1815

D
dzhwinter 已提交
1816 1817
    @property
    def random_seed(self):
Y
yuyang18 已提交
1818 1819 1820 1821 1822 1823
        """
        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
1824 1825
        return self._seed

Q
qiaolongfei 已提交
1826 1827
    @property
    def num_blocks(self):
Y
yuyang18 已提交
1828 1829 1830
        """
        The number of blocks in this program.
        """
Q
qiaolongfei 已提交
1831 1832
        return self.desc.num_blocks()

D
dzhwinter 已提交
1833 1834 1835 1836 1837 1838
    @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
1839
    def __repr__(self):
1840
        return self.__str__()
1841

Y
Yu Yang 已提交
1842
    def global_block(self):
Y
yuyang18 已提交
1843 1844 1845
        """
        Get the first block of this program.
        """
Y
Yu Yang 已提交
1846 1847
        return self.blocks[0]

Q
Qiao Longfei 已提交
1848
    def block(self, index):
Y
yuyang18 已提交
1849 1850 1851 1852 1853 1854 1855 1856
        """
        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
1857 1858
        return self.blocks[index]

Y
Yu Yang 已提交
1859
    def current_block(self):
Y
yuyang18 已提交
1860 1861 1862 1863
        """
        Get the current block. The :code:`current` block is the block to append
        operators.
        """
Y
Yu Yang 已提交
1864 1865
        return self.blocks[self.current_block_idx]

W
Wu Yi 已提交
1866
    def _create_block(self, parent_idx=None):
Y
yuyang18 已提交
1867 1868 1869 1870 1871 1872 1873 1874 1875 1876
        """
        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
1877
        new_block_idx = len(self.blocks)
F
update  
fengjiayi 已提交
1878 1879 1880
        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
1881 1882 1883 1884
        self.current_block_idx = new_block_idx
        self.blocks.append(Block(self, self.current_block_idx))
        return self.current_block()

W
Wu Yi 已提交
1885
    def _rollback(self):
Y
yuyang18 已提交
1886 1887 1888 1889 1890
        """
        Exit a code block, i.e., roll back to the parent block.
        Returns:
            None
        """
Y
Yu Yang 已提交
1891 1892
        self.current_block_idx = self.current_block().parent_idx

W
Wu Yi 已提交
1893
    def _sync_with_cpp(self):
Y
yuyang18 已提交
1894 1895 1896 1897 1898 1899 1900 1901 1902 1903
        """
        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
1904 1905 1906
        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
1907
            block._sync_with_cpp()
Q
Qiao Longfei 已提交
1908

W
Wu Yi 已提交
1909
    def _copy_param_info_from(self, other):
1910
        """
1911
        Copy the information of parameters from other program.
D
dzhwinter 已提交
1912

Y
yuyang18 已提交
1913 1914 1915
        Notes: This is a very low level API. Users should not invoke it
        directly.

1916 1917 1918 1919 1920 1921 1922
        Args:
            other(Program): Other program

        Returns:
            None
        """
        if not isinstance(other, Program):
W
Wu Yi 已提交
1923
            raise TypeError("_copy_param_info_from should be invoked with "
1924 1925 1926
                            "Program")

        if len(self.blocks) != len(other.blocks):
W
Wu Yi 已提交
1927
            raise ValueError("_copy_param_info_from should be invoked with two "
1928
                             "program, with represent the same topology")
W
Wu Yi 已提交
1929
        self.global_block()._copy_param_info_from(other.global_block())
1930

W
Wu Yi 已提交
1931
    def _copy_data_info_from(self, other):
F
fengjiayi 已提交
1932 1933
        """
        Copy the information of data variables from other program.
D
dzhwinter 已提交
1934

Y
yuyang18 已提交
1935 1936 1937
        Notes: This is a very low level API. Users should not invoke it
        directly.

F
fengjiayi 已提交
1938 1939 1940 1941 1942 1943 1944
        Args:
            other(Program): Other program

        Returns:
            None
        """
        if not isinstance(other, Program):
W
Wu Yi 已提交
1945
            raise TypeError("_copy_param_info_from should be invoked with "
F
fengjiayi 已提交
1946 1947 1948
                            "Program")

        if len(self.blocks) != len(other.blocks):
W
Wu Yi 已提交
1949
            raise ValueError("_copy_param_info_from should be invoked with two "
F
fengjiayi 已提交
1950
                             "program, with represent the same topology")
1951
        for var in list(other.global_block().vars.values()):
F
fengjiayi 已提交
1952 1953 1954
            if var.is_data:
                self.global_block().var(var.name).is_data = True

1955
    def list_vars(self):
Y
yuyang18 已提交
1956 1957 1958 1959 1960 1961
        """
        Get all variables from this Program. A iterable object is returned.

        Returns:
            iterable: The generator will yield every variable in this program.
        """
1962
        for each_block in self.blocks:
1963
            for each_var in list(each_block.vars.values()):
1964 1965
                yield each_var

Y
Yu Yang 已提交
1966

Y
Yu Yang 已提交
1967
class Parameter(Variable):
1968
    """
1969
    Parameter is derived from Variable. A parameter is a persistable
1970
    Variable, and will be updated by optimizers after each iteration.
1971
    The training of a neural network is essentially the updating of
1972 1973
    its parameters.

1974
    Relative to a general Variable, a Parameter has several its own
1975 1976
    member variables:

1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988
    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.
1989 1990
    """

Y
Yu Yang 已提交
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000
    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")
2001 2002 2003

        Variable.__init__(
            self, block, persistable=True, shape=shape, dtype=dtype, **kwargs)
Y
Yu Yang 已提交
2004 2005 2006 2007
        self.trainable = kwargs.get('trainable', True)

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

2008 2009
        self.regularizer = kwargs.get('regularizer', None)

F
fengjiayi 已提交
2010
        self.gradient_clip_attr = kwargs.get('gradient_clip_attr', None)
Y
Yu Yang 已提交
2011

W
wanghaoshuang 已提交
2012
        self.do_model_average = kwargs.get('do_model_average', None)
W
wanghaoshuang 已提交
2013

F
fengjiayi 已提交
2014 2015 2016
    def __str__(self):
        return self.to_string(True)

F
update  
fengjiayi 已提交
2017 2018 2019
    def to_string(self, throw_on_error, with_details=False):
        """
        To debug string.
D
dzhwinter 已提交
2020

F
update  
fengjiayi 已提交
2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034
        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
2035
                               "gradient_clip_attr", "do_model_average")
F
update  
fengjiayi 已提交
2036
            for attr_name in additional_attr:
2037 2038
                res_str += "%s: %s\n" % (
                    attr_name, six.binary_type(getattr(self, attr_name)))
F
update  
fengjiayi 已提交
2039 2040
        else:
            res_str = Variable.to_string(self, throw_on_error, False)
F
fengjiayi 已提交
2041 2042 2043 2044
        return res_str

    __repr__ = __str__

Y
Yu Yang 已提交
2045

Y
Yu Yang 已提交
2046
# program is a global instance.
Y
Yu Yang 已提交
2047 2048
_main_program_ = Program()
_startup_program_ = Program()
2049

2050

2051
def default_startup_program():
Y
Yu Yang 已提交
2052
    """
Y
yuyang18 已提交
2053 2054 2055 2056 2057 2058 2059 2060 2061
    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.
2062

Y
Yu Yang 已提交
2063 2064 2065
    Returns:
        Program: startup program
    """
Y
Yu Yang 已提交
2066
    return _startup_program_
2067

2068

2069
def default_main_program():
Y
Yu Yang 已提交
2070
    """
Y
yuyang18 已提交
2071 2072 2073 2074 2075 2076 2077 2078 2079
    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.
2080

Y
Yu Yang 已提交
2081 2082 2083
    Returns:
        Program: main program
    """
Y
Yu Yang 已提交
2084
    return _main_program_
Y
Yu Yang 已提交
2085 2086 2087 2088 2089


def switch_main_program(program):
    """
    Switch the main program to a new program.
2090

Y
Yu Yang 已提交
2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104
    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):
    """
2105
    Switch the startup program to a new program
Y
Yu Yang 已提交
2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120
    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
2121 2122 2123
    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.
2124

Y
Yu Yang 已提交
2125
    Examples:
Y
yuyang18 已提交
2126 2127 2128 2129 2130 2131 2132 2133 2134 2135

        >>> 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.
2136

Y
Yu Yang 已提交
2137
    Examples:
Y
yuyang18 已提交
2138 2139 2140 2141 2142 2143

        >>> 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 = ...
2144

Y
Yu Yang 已提交
2145
    Args:
Y
yuyang18 已提交
2146
        main_program(Program): New main program inside `with` statement.
2147
        startup_program(Program): New startup program inside `with` statement.
Y
Yu Yang 已提交
2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160
            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
2161 2162 2163 2164


def get_var(name, program=None):
    """
Y
yuyang18 已提交
2165
    Get a variable by name from the global block of a program.
F
fengjiayi 已提交
2166

X
xuwei06 已提交
2167 2168 2169
    Args:
        name(str): name of the variable
        program(Program|None): program object.
T
tangwei12 已提交
2170
        If None, default_global_program() will be used.
X
xuwei06 已提交
2171 2172 2173 2174 2175 2176 2177

    Returns:
        Variable
    """
    if program is None:
        program = default_main_program()
    assert isinstance(name, str)
2178
    assert isinstance(program, Program)
X
xuwei06 已提交
2179 2180

    return program.global_block().var(name)