optimizer.py 150.6 KB
Newer Older
1
# Copyright (c) 2019 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
# 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.
14 15

from __future__ import print_function
16

17
import numpy as np
18
from collections import defaultdict
19

Q
Qiao Longfei 已提交
20
from paddle.fluid.distribute_lookup_table import find_distributed_lookup_table
21
from paddle.fluid.framework import Program, Variable, name_scope, default_main_program, default_startup_program
22

23 24
from . import framework
from . import layers
25
from . import unique_name
H
hutuxian 已提交
26
from .backward import append_backward, _some_in_set_, _append_grad_suffix_
27
from .clip import append_gradient_clip_ops, error_clip_callback
28 29 30
from .framework import program_guard
from .initializer import Constant
from .layer_helper import LayerHelper
S
sneaxiy 已提交
31
from .layers import ops
32
from .regularizer import append_regularization_ops
33 34
from .dygraph import base as imperative_base
from .dygraph.learning_rate_scheduler import LearningRateDecay
H
hong 已提交
35
from .framework import _var_base_to_np
36 37 38
from paddle.fluid import core
from paddle.fluid.layers import tensor
from functools import reduce
39
from .wrapped_decorator import signature_safe_contextmanager
M
mapingshuo 已提交
40
from .. import compat as cpt
41

42
__all__ = [
43 44 45 46
    'SGD', 'Momentum', 'Adagrad', 'Adam', 'Adamax', 'Dpsgd', 'DecayedAdagrad',
    'Ftrl', 'SGDOptimizer', 'MomentumOptimizer', 'AdagradOptimizer',
    'AdamOptimizer', 'AdamaxOptimizer', 'DpsgdOptimizer',
    'DecayedAdagradOptimizer', 'RMSPropOptimizer', 'FtrlOptimizer', 'Adadelta',
Z
Zeng Jinle 已提交
47 48 49 50
    'AdadeltaOptimizer', 'ModelAverage', 'LarsMomentum',
    'LarsMomentumOptimizer', 'DGCMomentumOptimizer', 'LambOptimizer',
    'ExponentialMovingAverage', 'PipelineOptimizer', 'LookaheadOptimizer',
    'RecomputeOptimizer'
51
]
Q
Qiao Longfei 已提交
52 53 54 55 56 57


class Optimizer(object):
    """Optimizer Base class.

    Define the common interface of an optimizer.
58 59
    User should not use this class directly,
    but need to use one of it's implementation.
Q
Qiao Longfei 已提交
60 61
    """

62
    @imperative_base.no_grad
X
Xin Pan 已提交
63
    def __init__(self, learning_rate, regularization=None, name=None):
L
lujun 已提交
64
        if framework.in_dygraph_mode():
M
minqiyang 已提交
65 66 67 68 69
            if not isinstance(learning_rate, float) and \
                    not isinstance(learning_rate, LearningRateDecay):
                raise TypeError(
                    "learning rate should be float or LearningRateDecay, got %s here"
                    % type(learning_rate))
70 71 72 73
            if name is not None:
                self._name = unique_name.generate(name)
            else:
                self._name = unique_name.generate(self.__class__.__name__)
M
minqiyang 已提交
74 75 76 77 78 79
        else:
            if not isinstance(learning_rate, float) and \
                    not isinstance(learning_rate, framework.Variable):
                raise TypeError(
                    "learning rate should be float or Variable, got %s here" %
                    type(learning_rate))
80
            self._name = name
M
minqiyang 已提交
81

D
dzhwinter 已提交
82
        self.regularization = regularization
83
        self._learning_rate = learning_rate
D
dzhwinter 已提交
84 85
        # the learning rate type should be inferenced from loss
        self._dtype = None
86
        # each program should have a independent learning rate
87
        # program -> Variable(learning_rate)
Q
qiaolongfei 已提交
88
        self._learning_rate_map = dict()
89
        if isinstance(self._learning_rate, framework.Variable):
90 91
            self._learning_rate_map[framework.default_main_program(
            )] = self._learning_rate
92 93 94 95 96
        # Dictionary of accumulators. Some optimizer subclasses need to
        # allocate and manage extra variables associated with the parameters
        # to train. These variables are called accumulators.
        # {accum_name : { paramter_name : accumulator_for_parameter, ...}, ...}
        self._accumulators = defaultdict(lambda: dict())
Q
Qiao Longfei 已提交
97
        self.helper = None
98
        self._opti_name_list = []
H
hong 已提交
99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143
        self._accumulators_holder = {}

    @framework.dygraph_only
    def state_dict(self):
        '''
        Get state dict information from optimizer. It contain all the variable used by optimizer. For Adam opimizer, contains beta1, beta2, momentum etc. If LearningRateDecay have been used, global_step will be include in state dict.
        If the optimzier never be called(minimize function), the state_dict is empty.

        Args: None
        Return:
            state_dict(dict) : dict contains all the variablel used by optimizer
        
        Examples:
            .. code-block:: python

                import paddle.fluid as fluid
                adam = fluid.optimizer.Adam(0.001)
                state_dict = adam.state_dict()

        '''
        state_dict = {}
        for k, v in self._accumulators.items():
            for para_name, var_tmp in v.items():
                state_dict[var_tmp.name] = var_tmp
        # global step if use lr decay
        if isinstance(self._learning_rate, LearningRateDecay):
            var_temp = Variable(None, name='global_step', dtype='int32')
            tensor.fill_constant(
                [1], "int32", self._learning_rate.step_num, out=var_temp)

            state_dict['global_step'] = var_temp
        return state_dict

    @framework.dygraph_only
    def set_dict(self, state_dict):
        '''
        Load optimizer state dict. For Adam opimizer, contains beta1, beta2, momentum etc. If LearningRateDecay have been used, global_step will be changed.

        Args: 
            state_dict(dict) : Dict contains all the Variable needed by optimizer
        Return:
            None
        
        Examples:
            .. code-block:: python
144

H
hong 已提交
145 146
                with fluid.dygraph.guard():
                    emb = fluid.dygraph.Embedding( "emb", [10, 10])
147

H
hong 已提交
148 149
                    state_dict = emb.state_dict()
                    fluid.save_dygraph( state_dict, "paddle_dy")
150

H
hong 已提交
151 152 153
                    adam = fluid.optimizer.Adam( learning_rate = fluid.layers.noam_decay( 100, 10000) )
                    state_dict = adam.state_dict()
                    fluid.save_dygraph( state_dict, "padle_dy")
154

H
hong 已提交
155
                    para_state_dict, opti_state_dict = fluid.load_dygraph( "paddle_dy")
156

H
hong 已提交
157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216
                    adam.set_dict( opti_state_dict )

        '''

        if isinstance(self._learning_rate, LearningRateDecay):
            assert 'global_step' in state_dict, \
                    'Global step not in state dict, Dygraph use LearningRateDecay, global_step must in state_dict'
            global_step = state_dict['global_step']

            if isinstance(global_step, core.VarBase):
                step_np = global_step._copy_to(core.CPUPlace(), True)
                step_np = np.array(step_np.value().get_tensor())
                assert step_np.shape == (1,),  \
                        "global step shape is (1,), the shape is {}".format( step_np.shape )

                self._learning_rate.step_num = int(step_np[0])
            elif isinstance(global_step, Variable):
                step_np = global_step.numpy()
                assert step_np.shape == (1,),  \
                        "global step shape is (1,), the shape is {}".format( step_np.shape )
                self._learning_rate.step_num = step_np[0]
            elif isinstance(global_step, np.ndarray):
                assert global_step.shape == (1,),  \
                        "global step shape is (1,), the shape is {}".format( global_step.shape )
                self._learning_rate.step_num = global_step[0]
            else:
                raise RuntimeError(
                    "Type not supprt, value in state dict must be [VarBase, Variable, numpy], the type is ",
                    type(global_step))

        self._accumulators_holder = state_dict
        for k, v in self._accumulators.items():
            for para_name, var_tmp in v.items():
                assert var_tmp.name in state_dict, \
                        "optimizer variable {} not found".format( var_tmp.name )
                var = var_tmp._ivar.value()
                tensor = var.get_tensor()
                model_np = np.array(tensor)

                load_para = state_dict[var_tmp.name]

                if isinstance(load_para, Variable):
                    load_para_np = load_para.numpy()
                elif isinstance(load_para, core.VarBase):
                    load_para_np = _var_base_to_np(load_para)
                elif isinstance(load_para, np.ndarray):
                    load_para_np = load_para
                else:
                    raise RuntimeError("State dict type {} not supprt".format(
                        str(type(load_para))))

                assert model_np.shape == load_para_np.shape,  \
                                          "Parameter shape not match, Dygraph Parameter [ {} ] need tensor with shape {} but load tensor with shape {}".format(
                                                 item.name, model_np.shape, load_para_np.shape)

                assert model_np.dtype == load_para_np.dtype, \
                                          "Parameter dtype not match, Dygraph Parameter [ {} ] need tensor with dtype {}  but load tensor with dtype {}".format(
                                                item.name, model_np.dtype, load_para_np.dtype)

                tensor.set(load_para_np, framework._current_expected_place())
217

218 219
    def get_opti_var_name_list(self):
        return self._opti_name_list
Q
Qiao Longfei 已提交
220

Q
Qiao Longfei 已提交
221
    def _create_global_learning_rate(self):
222 223 224
        if imperative_base.enabled():
            # create learning rate Variable
            if isinstance(self._learning_rate, float):
M
minqiyang 已提交
225 226 227 228 229 230 231 232 233 234 235 236
                lr = self._global_learning_rate()

                if isinstance(lr, framework.Variable):
                    return
                else:
                    self._learning_rate_map[framework.default_main_program(
                    )] = layers.create_global_var(
                        name=unique_name.generate("learning_rate"),
                        shape=[1],
                        value=float(self._learning_rate),
                        dtype='float32' if self._dtype is None else self._dtype,
                        persistable=True)
237
            # get learning rate Variable from LearningRateDecay
M
minqiyang 已提交
238
            elif isinstance(self._learning_rate, LearningRateDecay):
239 240 241
                self._learning_rate_map[framework.default_main_program(
                )] = self._learning_rate()
            else:
Q
qiaolongfei 已提交
242
                raise TypeError(
243 244
                    "optimizer's learning rate must be float or LearningRateDecay"
                )
245
        else:
246 247 248 249
            lr = self._global_learning_rate()

            if isinstance(lr, framework.Variable):
                return
M
minqiyang 已提交
250 251 252 253 254 255
            else:
                if not isinstance(self._learning_rate, float):
                    raise TypeError(
                        "learning rate variable is create outside optimizer,"
                        "can not create new learning rate variable for new program"
                    )
Q
Qiao Longfei 已提交
256

257 258 259 260 261 262 263 264
            # create learning rate in the current main program
            self._learning_rate_map[framework.default_main_program(
            )] = layers.create_global_var(
                name=unique_name.generate("learning_rate"),
                shape=[1],
                value=float(self._learning_rate),
                dtype='float32' if self._dtype is None else self._dtype,
                persistable=True)
265

Y
yuyang18 已提交
266
    def _global_learning_rate(self, program=None):
Q
Qiao Longfei 已提交
267 268 269 270
        """
        get global decayed learning rate
        :return:
        """
271 272
        if program is None:
            program = framework.default_main_program()
Q
qiaolongfei 已提交
273
        return self._learning_rate_map.get(program, None)
Q
Qiao Longfei 已提交
274

Q
Qiao Longfei 已提交
275 276 277 278 279
    def _append_optimize_op(self, block, param_and_grad):
        """ append optimize operator to block and return all the added optimize_op
        """
        raise NotImplementedError()

280 281 282 283
    def _create_param_lr(self, param_and_grad):
        # create learning rate variable for every parameter
        param = param_and_grad[0]
        param_lr = param.optimize_attr['learning_rate']
W
Wu Yi 已提交
284 285
        if type(param_lr) == Variable:
            return param_lr
Q
qiaolongfei 已提交
286
        else:
W
Wu Yi 已提交
287
            if param_lr == 1.0:
Y
yuyang18 已提交
288
                return self._global_learning_rate()
W
Wu Yi 已提交
289
            else:
X
Xin Pan 已提交
290 291 292
                with default_main_program()._lr_schedule_guard(
                        is_with_opt=True), framework.name_scope(
                            'scale_with_param_lr'):
293
                    return self._global_learning_rate() * param_lr
294 295 296 297 298 299 300

    def _create_accumulators(self, block, parameters):
        """Create all accumulators needed by the parameters

        Args:
            block: the block in which the loss variable is present
            parameters: list of parameter variables for the optimizer
Q
Qiao Longfei 已提交
301
        """
302 303
        pass

304
    def _finish_update(self, block, parameters_and_grads):
305 306 307 308 309 310 311 312
        """Finish any custom updates needed
           before completing an optimization step

        Args:
            block: the block in which the loss variable is present
            parameters: list of parameter variables for the optimizer

        Returns:
Q
qiaolongfei 已提交
313
            None
314 315 316
        """
        pass

317 318 319 320 321 322
    def _add_accumulator(self,
                         name,
                         param,
                         dtype=None,
                         fill_value=0.0,
                         shape=None):
323 324 325 326 327 328 329 330 331
        """Utility function to add an accumulator for a parameter

        Args:
            block: the block in which the loss variable is present
            name: name of the accumulator
            param: parameter variable for which accumulator is to be added
            dtype: data type of the accumulator variable
            fill_value: value to initialize the accumulator variable
        """
W
whs 已提交
332 333
        if self._name is not None:
            name = self._name + "_" + name
334 335
        if (name in self._accumulators and
                param.name in self._accumulators[name]):
L
lujun 已提交
336
            if framework.in_dygraph_mode():
X
polish  
Xin Pan 已提交
337
                return self._accumulators[name][param.name]
338
            raise Exception("Accumulator {} already exists for parameter {}".
339
                            format(name, param.name))
340 341
        if shape == None:
            shape = param.shape
Q
Qiao Longfei 已提交
342
        assert isinstance(self.helper, LayerHelper)
343 344 345 346 347

        var_name = param.name + "_" + name
        var_name = unique_name.generate(var_name)
        self._opti_name_list.append(var_name)

Q
Qiao Longfei 已提交
348
        var = self.helper.create_global_variable(
349
            name=var_name,
Q
Qiao Longfei 已提交
350
            persistable=True,
F
fengjiayi 已提交
351
            dtype=dtype or param.dtype,
Q
Qiao Longfei 已提交
352
            type=param.type,
H
hong 已提交
353 354
            shape=shape,
            belong_to_optimizer=True)
Q
Qiao Longfei 已提交
355
        self.helper.set_variable_initializer(
356
            var, initializer=Constant(value=float(fill_value)))
H
hong 已提交
357 358 359 360 361 362 363

        if framework.in_dygraph_mode():
            if len(self._accumulators_holder) > 0:
                assert var_name in self._accumulators_holder, \
                        "Optimizer set error, {} should in state dict".format( var_name )
                var.set_value(self._accumulators_holder[var_name])

Q
Qiao Longfei 已提交
364
        self._accumulators[name][param.name] = var
365
        return var
366 367 368 369 370 371 372 373 374 375 376

    def _get_accumulator(self, name, param):
        """Utility function to fetch an accumulator for a parameter

        Args:
            name: name of the accumulator
            param: parameter variable for which accumulator is to be fetched

        Returns:
            accumulator variable for the parameter
        """
W
whs 已提交
377 378
        if self._name is not None:
            name = self._name + "_" + name
379 380 381 382 383 384
        if (name not in self._accumulators or
                param.name not in self._accumulators[name]):
            raise Exception("Accumulator {} does not exist for parameter {}".
                            format(name, param.name))
        return self._accumulators[name][param.name]

385
    def _create_optimization_pass(self, parameters_and_grads):
Q
Qiao Longfei 已提交
386 387 388
        """Add optimization operators to update gradients to variables.

        Args:
Q
qiaolongfei 已提交
389
          parameters_and_grads(list(tuple(Variable, Variable))):
390
            a list of (variable, gradient) pair to update.
Q
Qiao Longfei 已提交
391 392

        Returns:
393
          return_op_list: a list of operators that will complete one step of
394 395 396
            optimization. This will include parameter update ops, global step
            update ops and any other custom ops required by subclasses to manage
            their internal state.
Q
Qiao Longfei 已提交
397
        """
398 399 400 401 402
        # This is a default implementation of create_optimization_pass that
        # can be shared by most optimizers. This implementation assumes that
        # the subclass will implement the _append_optimize_op method and the
        #  _initialize_tensors method. The subclass can extend the
        # _create_accumulators method if it needs to create accumulators
403
        # for parameters and extend _finish_update method to add custom ops.
404

405 406 407 408
        # Allways called under program_guard use global block as loss block
        global_block = framework.default_main_program().global_block()
        start = len(global_block.ops)
        self.helper = LayerHelper(self.__class__.__name__)
C
chengduo 已提交
409 410 411
        self._create_accumulators(
            global_block,
            [p[0] for p in parameters_and_grads if p[0].trainable])
412 413 414
        self._create_global_learning_rate()

        optimize_ops = []
M
minqiyang 已提交
415
        if framework.in_dygraph_mode():
416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434
            for param_and_grad in parameters_and_grads:
                if param_and_grad[1] is None:
                    continue
                with param_and_grad[0].block.program._optimized_guard(
                        param_and_grad):
                    if param_and_grad[0].trainable is True:
                        optimize_op = self._append_optimize_op(global_block,
                                                               param_and_grad)
                        optimize_ops.append(optimize_op)
        else:
            for param_and_grad in parameters_and_grads:
                if param_and_grad[1] is None:
                    continue
                with param_and_grad[0].block.program._optimized_guard(
                        param_and_grad), name_scope("optimizer"):
                    if param_and_grad[0].trainable is True:
                        optimize_op = self._append_optimize_op(global_block,
                                                               param_and_grad)
                        optimize_ops.append(optimize_op)
435 436 437 438 439 440 441 442 443

        # Get custom finish ops for subclasses
        # FIXME: Need to fix this once we figure out how to handle dependencies
        self._finish_update(global_block, parameters_and_grads)

        end = len(global_block.ops)
        return global_block._slice_ops(start, end)

    def _process_distribute_lookuptable(self, param_grads):
Q
Qiao Longfei 已提交
444 445 446 447 448 449 450 451 452
        """
        Because distribute lookup table only support SGD optimizer for now, not support
        other optimizer and regularization, so we should find the table parameter out,
        and avoid to add regularization and other op for it, and add sgd optimize op
        for it independently.
        :param param_grads(list((Var, Var))): list of (param, grad) pair.
        :param loss: the loss variable.
        :param startup_program: the startup program
        """
453 454
        program = framework.default_main_program()
        global_block = framework.default_main_program().global_block()
Q
Qiao Longfei 已提交
455 456 457 458 459 460 461 462 463 464 465 466 467 468 469
        table_name = find_distributed_lookup_table(program)
        table_param = None
        table_grad = None
        new_param_grads = []
        for p, g in param_grads:
            if p.name == table_name:
                if table_param is not None:
                    raise RuntimeError(
                        "multi dist table var found, only support one now!")
                table_param = p
                table_grad = g
            else:
                new_param_grads.append((p, g))
        sgd_op = None
        if table_param is not None:
470 471 472 473 474 475 476 477 478 479 480 481 482
            param_and_grad = [table_param, table_grad]
            with table_param.block.program._optimized_guard(param_and_grad), \
                    framework.name_scope("optimizer"):
                self._create_global_learning_rate()
                # create the optimize op
                sgd_op = global_block.append_op(
                    type='sgd',
                    inputs={
                        "Param": table_param,
                        "Grad": table_grad,
                        "LearningRate": self._create_param_lr(param_and_grad)
                    },
                    outputs={"ParamOut": param_and_grad[0]})
Q
Qiao Longfei 已提交
483 484
        return new_param_grads, (table_param, table_grad), sgd_op

485 486 487
    def _append_dgc_ops(self, param_and_grad):
        pass

488 489 490 491 492 493 494
    def backward(self,
                 loss,
                 startup_program=None,
                 parameter_list=None,
                 no_grad_set=None,
                 callbacks=None):
        """
495
        The first part of ``minimize``, do auto-diff to append backward operations for
496 497 498
        the current program.

        Args:
499 500 501 502 503 504 505 506 507 508 509
            loss (Variable): ``loss`` variable to run optimizations.
            startup_program (Program, optional): :ref:`api_fluid_Program` for
                initializing parameters in ``parameter_list``. The default value
                is None, at this time :ref:`api_fluid_default_startup_program` will be used.
            parameter_list (list, optional): List of ``Variable`` names to update
                to minimize ``loss``. The default value is None, at this time all parameters
                will be updated.
            no_grad_set (set, optional): Set of ``Variable`` objects that don't need
                to be updated. The default value is None.
            callbacks (list, optional): list of callable objects to run when appending backward
                operator for one parameter. The default value is None.
M
minqiyang 已提交
510

511
        Return:
512 513
            list: list of (param, grad) variable pairs, param is ``Parameter``,
                grad is the gradient value corresponding to the parameter.
M
minqiyang 已提交
514

515
        Examples:
516
            See examples in ``apply_gradients``.
517
        """
G
gongweibao 已提交
518 519
        no_grad_set = self._get_no_grad_set(loss, no_grad_set)

C
chengduo 已提交
520
        self._dtype = loss.dtype
L
lujun 已提交
521
        if framework.in_dygraph_mode():
C
chengduo 已提交
522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538
            if parameter_list is not None:
                parameters = parameter_list
            else:
                parameters = framework._dygraph_tracer().all_parameters()

            params_grads = []
            for param in parameters:
                if not param.trainable:
                    continue
                if param._ivar._grad_ivar() is not None:
                    # create gradient variable
                    grad_var = Variable(
                        block=loss.block,
                        name=param._ivar._grad_name(),
                        stop_gradient=True,
                        ivar=param._ivar._grad_ivar())
                    params_grads.append((param, grad_var))
539
        else:
C
chengduo 已提交
540 541 542 543 544
            if callbacks is None:
                callbacks = [error_clip_callback]
            else:
                assert (isinstance(callbacks, list))
            program = loss.block.program
C
chengduo 已提交
545 546 547 548
            assert len(loss.shape) == 1 and loss.shape[0] == 1, \
                "The loss.shape should be (1L,), but the current loss.shape is {}. " \
                "Maybe that you should call fluid.layers.mean to process the current loss.".format(
                    loss.shape)
C
chengduo 已提交
549 550 551 552 553 554 555
            with program_guard(program, startup_program):
                params_grads = append_backward(loss, parameter_list,
                                               no_grad_set, callbacks)
                # Note: since we can't use all_reduce_op now,
                #  dgc_op should be the last op of one grad.
                self._append_dgc_ops(params_grads)
        return params_grads
556 557 558 559 560 561 562 563

    def apply_gradients(self, params_grads):
        """
        Second part of `minimize`, appending optimization operators for
        given `params_grads` pairs.

        Args:
            params_grads (list): list of (param, grad) pair to do optimization.
M
minqiyang 已提交
564

565 566
        Returns:
            list: A list of operators appended to the current program.
M
minqiyang 已提交
567

568 569 570
        Examples:
            .. code-block:: python

571
                import paddle.fluid as fluid
572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596
                loss = network()
                optimizer = fluid.optimizer.SGD(learning_rate=0.1)
                params_grads = optimizer.backward(loss)
                # you may append operations for params_grads here
                # ...
                optimizer.apply_gradients(params_grads)
        """
        params_grads = sorted(params_grads, key=lambda x: x[0].name)

        params_grads, table_param_and_grad, table_optimize_op = \
            self._process_distribute_lookuptable(params_grads)

        params_grads = append_gradient_clip_ops(params_grads)

        # Add regularization if any
        params_grads = append_regularization_ops(params_grads,
                                                 self.regularization)

        optimize_ops = self._create_optimization_pass(params_grads)
        if table_optimize_op is not None:
            optimize_ops.append(table_optimize_op)
            params_grads.append(table_param_and_grad)

        return optimize_ops

C
chengduo 已提交
597 598 599 600 601 602 603 604 605 606 607 608 609 610
    def apply_optimize(self, loss, startup_program, params_grads):
        """
        Second part of `minimize`, appending optimization operators for
        given `params_grads` pairs.

        Args:
            loss (Variable): loss variable to run optimizations.
            startup_program (Program): startup_program for initializing parameters
                in `parameter_list`.
            params_grads (list): list of (param, grad) pair to do optimization.

        Returns:
            list: A list of operators appended to the current program.
        """
L
lujun 已提交
611
        if framework.in_dygraph_mode():
C
chengduo 已提交
612 613
            with program_guard(framework.default_main_program(),
                               framework.default_startup_program()):
614 615
                params_grads = append_regularization_ops(params_grads,
                                                         self.regularization)
C
chengduo 已提交
616 617 618 619 620 621 622
                optimize_ops = self._create_optimization_pass(params_grads)
        else:
            program = loss.block.program
            with program_guard(program, startup_program):
                optimize_ops = self.apply_gradients(params_grads)
        return optimize_ops

G
gongweibao 已提交
623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639
    def _get_no_grad_set(self, loss, no_grad_set=None):
        if no_grad_set is None:
            no_grad_set = set()
        elif isinstance(no_grad_set, set) or isinstance(
                no_grad_set, list) or isinstance(no_grad_set, tuple):
            no_grad_set = set(no_grad_set)
        else:
            assert "no_grad_set should be a set, but the passed type is {}".format(
                type(no_grad_set))
        parameters = loss.block.program.global_block().all_parameters()
        param_no_trainable = set(
            [param.name for param in parameters if param.trainable is False])
        # If the parameter is no trainable, it should not have a gradient.
        no_grad_set.update(param_no_trainable)

        return no_grad_set

640
    @imperative_base.no_grad
Q
Qiao Longfei 已提交
641 642
    def minimize(self,
                 loss,
643
                 startup_program=None,
Q
Qiao Longfei 已提交
644
                 parameter_list=None,
645 646
                 no_grad_set=None,
                 grad_clip=None):
647
        """
648
        Add operations to minimize ``loss`` by updating ``parameter_list``.
M
minqiyang 已提交
649

650
        Args:
651 652 653 654 655 656 657 658 659 660 661 662 663
            loss (Variable): A ``Variable`` containing the value to minimize.
            startup_program (Program, optional): :ref:`api_fluid_Program` for
                initializing parameters in ``parameter_list``. The default value
                is None, at this time :ref:`api_fluid_default_startup_program` will be used.
            parameter_list (list, optional): List of ``Variable`` names to update
                to minimize ``loss``. The default value is None, at this time all parameters
                will be updated.
            no_grad_set (set, optional): Set of ``Variable`` objects that don't need
                to be updated. The default value is None.
            grad_clip (GradClipBase, optional) : Gradient clipping strategy, static
                graph mode does not need to use this argument. Currently, this argument
                only supports gradient clipping in dygraph mode. In the future, this
                argument my be adjusted. The default value is None.
Q
Qiao Longfei 已提交
664

665
        Returns:
666 667 668 669 670 671
            tuple: tuple (optimize_ops, params_grads), A list of operators appended
            by minimize and a list of (param, grad) variable pairs, param is
            ``Parameter``, grad is the gradient value corresponding to the parameter.

        Examples:
            Please refer to the example of current Optimizer.
Q
Qiao Longfei 已提交
672
        """
C
chengduo 已提交
673
        assert isinstance(loss, Variable), "The loss should be an Variable."
C
chengduo 已提交
674 675 676 677 678
        params_grads = self.backward(
            loss,
            startup_program=startup_program,
            parameter_list=parameter_list,
            no_grad_set=no_grad_set)
679 680 681 682 683

        if grad_clip is not None and framework.in_dygraph_mode():
            # TODO(hongyu): FIX later, this is only for dygraph, should be work for static mode
            params_grads = grad_clip(params_grads)

C
chengduo 已提交
684 685
        optimize_ops = self.apply_optimize(
            loss, startup_program=startup_program, params_grads=params_grads)
M
minqiyang 已提交
686

Q
Qiao Longfei 已提交
687
        return optimize_ops, params_grads
Q
Qiao Longfei 已提交
688 689 690


class SGDOptimizer(Optimizer):
Q
qiaolongfei 已提交
691 692 693 694 695 696 697
    """
    Optimizer of the stochastic gradient descent algorithm.

    .. math::

        param\_out = param - learning\_rate * grad

698 699 700 701 702 703 704
    Parameters:
        learning_rate (float|Variable): The learning rate used to update parameters. \
            Can be a float value or a Variable with one float value as data element.
        regularization: A Regularizer, such as :ref:`api_fluid_regularizer_L2DecayRegularizer`. \
            Optional, default is None.
        name (str, optional): This parameter is used by developers to print debugging information. \
            For details, please refer to :ref:`api_guide_Name`. Default is None.
Q
qiaolongfei 已提交
705 706 707 708

    Examples:
        .. code-block:: python

709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733
            import paddle
            import paddle.fluid as fluid
            import numpy as np

            place = fluid.CPUPlace()
            main = fluid.Program()
            with fluid.program_guard(main):
                x = fluid.layers.data(name='x', shape=[13], dtype='float32')
                y = fluid.layers.data(name='y', shape=[1], dtype='float32')
                y_predict = fluid.layers.fc(input=x, size=1, act=None)
                cost = fluid.layers.square_error_cost(input=y_predict, label=y)
                avg_cost = fluid.layers.mean(cost)

                sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.001)
                sgd_optimizer.minimize(avg_cost)

                fetch_list = [avg_cost]
                train_reader = paddle.batch(
                    paddle.dataset.uci_housing.train(), batch_size=1)
                feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
                exe = fluid.Executor(place)
                exe.run(fluid.default_startup_program())
                for data in train_reader():
                    exe.run(main, feed=feeder.feed(data), fetch_list=fetch_list)

Q
Qiao Longfei 已提交
734 735
    """

X
Xin Pan 已提交
736
    def __init__(self, learning_rate, regularization=None, name=None):
Q
Qiao Longfei 已提交
737
        assert learning_rate is not None
Q
Qiao Longfei 已提交
738
        super(SGDOptimizer, self).__init__(
X
Xin Pan 已提交
739 740 741
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
Q
Qiao Longfei 已提交
742 743
        self.type = "sgd"

744 745
    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)
746

Q
Qiao Longfei 已提交
747 748 749 750 751 752
        # create the optimize op
        sgd_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
753
                "LearningRate": self._create_param_lr(param_and_grad)
Q
Qiao Longfei 已提交
754
            },
M
minqiyang 已提交
755 756
            outputs={"ParamOut": param_and_grad[0]},
            stop_gradient=True)
Q
Qiao Longfei 已提交
757 758

        return sgd_op
759 760 761


class MomentumOptimizer(Optimizer):
Q
qiaolongfei 已提交
762 763 764 765 766 767 768 769 770 771 772 773 774 775
    """

    Simple Momentum optimizer with velocity state

    This optimizer has a flag for Nestrov Momentum.

    The update equations are as follows:

    .. math::

        & velocity = mu * velocity + gradient

        & if (use\_nesterov):

776
        &\quad   param = param - (gradient + mu * velocity) * learning\_rate
Q
qiaolongfei 已提交
777 778 779

        & else:

Q
qiaolongfei 已提交
780
        &\quad   param = param - learning\_rate * velocity
Q
qiaolongfei 已提交
781

782 783 784 785 786 787 788 789 790
    Parameters:
        learning_rate (float|Variable): The learning rate used to update parameters. \
            Can be a float value or a Variable with one float value as data element.
        momentum (float): Momentum factor
        use_nesterov (bool, optional): Enables Nesterov momentum, default is false.
        regularization: A Regularizer, such as :ref:`api_fluid_regularizer_L2DecayRegularizer`. \
            Optional, default is None.
        name (str, optional): This parameter is used by developers to print debugging information. \
            For details, please refer to :ref:`api_guide_Name`. Default is None.
Q
qiaolongfei 已提交
791 792 793 794

    Examples:
        .. code-block:: python

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
            import paddle
            import paddle.fluid as fluid
            import numpy as np

            place = fluid.CPUPlace()
            main = fluid.Program()
            with fluid.program_guard(main):
                x = fluid.layers.data(name='x', shape=[13], dtype='float32')
                y = fluid.layers.data(name='y', shape=[1], dtype='float32')
                y_predict = fluid.layers.fc(input=x, size=1, act=None)
                cost = fluid.layers.square_error_cost(input=y_predict, label=y)
                avg_cost = fluid.layers.mean(cost)

                moment_optimizer = fluid.optimizer.MomentumOptimizer(learning_rate=0.001, momentum=0.9)
                moment_optimizer.minimize(avg_cost)

                fetch_list = [avg_cost]
                train_reader = paddle.batch(
                    paddle.dataset.uci_housing.train(), batch_size=1)
                feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
                exe = fluid.Executor(place)
                exe.run(fluid.default_startup_program())
                for data in train_reader():
                    exe.run(main, feed=feeder.feed(data), fetch_list=fetch_list)

820 821 822
    """
    _velocity_acc_str = "velocity"

X
Xin Pan 已提交
823 824 825 826 827 828
    def __init__(self,
                 learning_rate,
                 momentum,
                 use_nesterov=False,
                 regularization=None,
                 name=None):
829 830
        assert learning_rate is not None
        assert momentum is not None
Q
Qiao Longfei 已提交
831
        super(MomentumOptimizer, self).__init__(
X
Xin Pan 已提交
832 833 834
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
835 836
        self.type = "momentum"
        self._momentum = momentum
837
        self._use_nesterov = bool(use_nesterov)
838 839 840 841 842

    def _create_accumulators(self, block, parameters):
        assert isinstance(block, framework.Block)

        for p in parameters:
Q
Qiao Longfei 已提交
843
            self._add_accumulator(self._velocity_acc_str, p)
844 845 846 847 848 849 850 851 852 853 854 855 856

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)

        velocity_acc = self._get_accumulator(self._velocity_acc_str,
                                             param_and_grad[0])
        # create the momentum optimize op
        momentum_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "Velocity": velocity_acc,
857
                "LearningRate": self._create_param_lr(param_and_grad)
858 859 860 861 862
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "VelocityOut": velocity_acc
            },
863
            attrs={"mu": self._momentum,
M
minqiyang 已提交
864 865
                   "use_nesterov": self._use_nesterov},
            stop_gradient=True)
866 867

        return momentum_op
868 869


870
class DGCMomentumOptimizer(Optimizer):
871
    """
872
    DGC (Deep Gradient Compression) Momentum Optimizer. Original paper is https://arxiv.org/abs/1712.01887
873

G
gongweibao 已提交
874
    DGC reduces the communication bandwidth by sending only the important gradients (sparse update):\
875 876
        only gradients larger than a threshold are transmitted.

G
gongweibao 已提交
877
    To avoid losing information, DGC accumulates the rest of the gradients locally.
878 879 880

    Eventually, these gradients become large enough to be transmitted.

881
    Thus, DGC sends the large gradients immediately but eventually sends all of the gradients over time.
882

G
gongweibao 已提交
883
    To ensure no loss of accuracy, DGC employs momentum correction and local gradient clipping on top of the gradient sparsification to maintain model performance.
884 885 886 887

    DGC also uses momentum factor masking and warmup training to overcome the staleness problem caused by reduced communication.

    This optimizer will do two things:
888

889 890
        1. Compress the gradient by get TopK import value from tensor \
            and use it for allreduce to reduce network bandwidth.
891

892
        2. Call momentum to optimize the cost.
893 894

    Args:
895 896
        learning_rate (float|Variable): The learning rate used to update parameters. \
            It can be a float value or a Variable with one float value as a data element.
897
        momentum (float): Momentum factor.
G
gongweibao 已提交
898
        rampup_begin_step (int): The beginning step from which gradient compression is implemented.
899 900 901 902 903 904 905 906 907 908 909 910 911 912
        rampup_step (int): Time steps used in sparsity warm-up periods. Default is 1.
            For example, if the sparsity is [0.75, 0.9375, 0.984375, 0.996, 0.999], and the rampup_step is 100, \
                it will use 0.75 at 0~19 steps, and 0.9375 at 20~39 steps, and so on. \
                And when reach sparsity array ends, it will use 0.999 then and after.
        sparsity (list[float]): Get top important element from gradient tensor, the ratio is (1 - current sparsity). \
            Default is [0.999]. For example, if the sparsity is [0.99, 0.999], \
                the top [1%, 0.1%] important element will be transmitted.
        use_nesterov (bool): Enables Nesterov momentum. True means use Nesterov. Default is False.
        local_grad_clip_norm (float, optional): Local gradient clip norm value. Optional, default is None, represent no need clip.
        num_trainers (int, optional): The number of training nodes. Optional, default is None.
        regularization (WeightDecayRegularizer, optional): A Regularizer, such as \
            :ref:`api_fluid_regularizer_L2DecayRegularizer`. Optional, default is None.
        name (str, optional): This parameter is used by developers to print debugging information. \
            For details, please refer to :ref:`api_guide_Name`. Default is None.
913 914 915 916

    Examples:
        .. code-block:: python

917
            import paddle.fluid as fluid
918
            optimizer = fluid.optimizer.DGCMomentumOptimizer(
G
gongweibao 已提交
919 920 921 922 923
                        learning_rate=0.0001,
                        momentum=0.9,
                        rampup_step=1000,
                        rampup_begin_step=1252,
                        sparsity=[0.999, 0.999])
924 925

    """
926 927
    _u_velocity_acc_str = "_dgc_u_"
    _v_velocity_acc_str = "_dgc_v_"
928 929 930 931 932 933 934 935 936 937 938 939

    def __init__(self,
                 learning_rate,
                 momentum,
                 rampup_begin_step,
                 rampup_step=1,
                 sparsity=[0.999],
                 use_nesterov=False,
                 local_grad_clip_norm=None,
                 num_trainers=None,
                 regularization=None,
                 name=None):
940 941 942 943 944 945 946 947 948
        assert learning_rate is not None
        assert momentum is not None
        super(DGCMomentumOptimizer, self).__init__(
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
        self.type = "dgc_momentum"
        self._momentum = momentum
        self._use_nesterov = bool(use_nesterov)
949 950

        self._rampup_begin_step = rampup_begin_step
951 952
        self._rampup_step = rampup_step
        self._sparsity = sparsity
953

954
        self._rampup_begin_step_var = None
955
        self._global_step_var = None
956

957 958 959 960 961 962 963 964 965 966 967 968
        self._local_grad_clip_norm = None
        self._clip_norm = None
        if local_grad_clip_norm is not None:
            assert isinstance(num_trainers, int)
            assert isinstance(local_grad_clip_norm, float)
            assert num_trainers > 0

            self._local_grad_clip_norm = local_grad_clip_norm
            self._num_trainers = num_trainers
            self._clip_norm = local_grad_clip_norm / (num_trainers *
                                                      num_trainers)

969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984
        self._get_dgc_regularization_param()

    def _get_dgc_regularization_param(self):
        self.regular_coeff = 0.0
        self.regular_type = 0

        if self.regularization is not None:
            self.regular_coeff = self.regularization._regularization_coeff
            from .regularizer import L1Decay, L2Decay
            if isinstance(self.regularization, L1Decay):
                self.regular_type = 1
            elif isinstance(self.regularization, L2Decay):
                self.regular_type = 2
            else:
                assert False, 'regularization must be None|L1Decay|L2Deacy'

985 986 987 988 989 990 991 992 993 994 995
    def _is_use_dgc(self, param_var, grad_var):
        var_numel = abs(reduce(lambda x, y: x * y, param_var.shape))
        if var_numel < 16384 or \
           param_var.type == core.VarDesc.VarType.SELECTED_ROWS  or \
           grad_var.type == core.VarDesc.VarType.SELECTED_ROWS  or  \
               param_var.dtype != core.VarDesc.VarType.FP32 :
            return False
        return True

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)
996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010
        velocity_acc = self._get_accumulator(self._u_velocity_acc_str,
                                             param_and_grad[0])
        assert velocity_acc is not None

        inputs = {
            "Param": param_and_grad[0],
            "Grad": param_and_grad[1],
            "Velocity": velocity_acc,
            "LearningRate": self._create_param_lr(param_and_grad),
        }
        outputs = {
            "ParamOut": param_and_grad[0],
            "VelocityOut": velocity_acc,
        }
        attrs = {"mu": self._momentum, "use_nesterov": self._use_nesterov}
1011 1012

        if not self._is_use_dgc(param_and_grad[0], param_and_grad[1]):
1013 1014 1015
            type = "momentum"
        else:
            type = "dgc_momentum"
1016 1017 1018 1019 1020
            inputs.update({
                "current_step": self._global_step_var,
                "nranks": self._nranks_var
            })
            outputs.update({'Grad_out': param_and_grad[1]})
1021
            attrs.update({"rampup_begin_step": float(self._rampup_begin_step)})
1022 1023 1024

        # create the dgc momentum optimize op
        dgc_momentum_op = block.append_op(
1025 1026 1027 1028
            type=type,
            inputs=inputs,
            outputs=outputs,
            attrs=attrs,
1029 1030 1031
            stop_gradient=True)
        return dgc_momentum_op

1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050
    def _add_auto_increment_var(self, counter_name, begin, step=1):
        helper = LayerHelper('global_step_counter')
        counter, is_new_var = helper.create_or_get_global_variable(
            name=counter_name, dtype='float32', shape=[1], persistable=True)
        if is_new_var:
            helper.set_variable_initializer(
                counter,
                initializer=Constant(
                    value=float(begin - 1), force_cpu=True))
            helper.main_program.global_block()._prepend_op(
                type='increment',
                inputs={'X': [counter]},
                outputs={'Out': [counter]},
                attrs={'step': float(step)},
                stop_gradient=True)
            counter.stop_gradient = True

        return counter

1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063
    def _add_nranks_var(self, name, value=-1):
        helper = LayerHelper('global_step_counter')
        counter, is_new_var = helper.create_or_get_global_variable(
            name=name, dtype='float32', shape=[1], persistable=True)
        if is_new_var:
            helper.set_variable_initializer(
                counter,
                initializer=Constant(
                    value=float(value), force_cpu=True))
            counter.stop_gradient = True

        return counter

1064 1065 1066 1067 1068 1069
    def _append_dgc_ops(self, param_and_grads):
        main_program = default_main_program()
        main_program._enable_dgc = True

        # step counter
        self._global_step_var = self._add_auto_increment_var(
G
gongweibao 已提交
1070
            counter_name=core.dgc.kDGCCounterName(), begin=0)
1071

1072 1073 1074
        self._nranks_var = self._add_nranks_var(
            name=core.dgc.kDGCNRanksName(), value=-1)

1075 1076 1077 1078 1079
        # rampup begin step var for all_reduce_op_handle
        self._rampup_begin_step_var = tensor.create_global_var(
            shape=[1],
            dtype=core.VarDesc.VarType.FP32,
            persistable=True,
G
gongweibao 已提交
1080
            name=core.dgc.kDGCRampUpBeginStepName(),
1081 1082 1083
            value=self._rampup_begin_step * 1.0,
            force_cpu=True)

1084 1085
        self.helper = LayerHelper(self.__class__.__name__)

1086
        for param_var, grad_var in param_and_grads:
1087 1088 1089
            # reuse velocity in dgc_op and dgc_momentum_op
            u_var = self._add_accumulator(self._u_velocity_acc_str, param_var)

1090
            if not self._is_use_dgc(param_var, grad_var):
1091 1092
                continue

1093
            v_var = self._add_accumulator(self._v_velocity_acc_str, param_var)
1094 1095 1096 1097 1098

            k_var = tensor.create_global_var(
                shape=[1],
                dtype=param_var.dtype,
                persistable=True,
G
gongweibao 已提交
1099
                name=param_var.name + core.dgc.kDGCKName(),
1100 1101 1102 1103 1104 1105 1106
                value=0.0,
                force_cpu=True)

            encoded_var = tensor.create_global_var(
                shape=[1],
                dtype=param_var.dtype,
                persistable=True,
G
gongweibao 已提交
1107
                name=param_var.name + core.dgc.kDGCEncodedName(),
1108 1109 1110
                value=0.0,
                force_cpu=False)

1111 1112 1113 1114 1115 1116 1117 1118
            gather_var = tensor.create_global_var(
                shape=[1],
                dtype=param_var.dtype,
                persistable=True,
                name=param_var.name + core.dgc.kDGCGatherName(),
                value=0.0,
                force_cpu=False)

1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140
            # del back oprolevarname
            op_maker = core.op_proto_and_checker_maker
            backward = core.op_proto_and_checker_maker.OpRole.Backward
            for op in main_program.global_block().ops:
                if not self._is_the_backward_op(op):
                    continue

                var_attr = op.all_attrs()[op_maker.kOpRoleVarAttrName()]
                if param_var.name not in var_attr:
                    continue

                var_attr.remove(param_var.name)
                var_attr.remove(grad_var.name)
                if len(var_attr) > 1:
                    op._set_attr(op_maker.kOpRoleVarAttrName(), var_attr)
                else:
                    op._remove_attr(op_maker.kOpRoleVarAttrName())

            clip_var = grad_var
            if self._local_grad_clip_norm is not None:
                clip_var = self._append_clip_norm(grad_var, self._clip_norm)
            self._dgc_op(param_var, clip_var, grad_var, u_var, v_var, k_var,
1141
                         encoded_var, gather_var)
1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156

    def _is_the_backward_op(self, op):
        op_maker = core.op_proto_and_checker_maker
        backward = core.op_proto_and_checker_maker.OpRole.Backward
        if op_maker.kOpRoleVarAttrName() in op.attr_names and \
                int(op.all_attrs()[op_maker.kOpRoleAttrName()]) == int(backward):
            return True
        return False

    def _clip_by_norm(self, x, max_norm, name=None):
        args = {'x': x, 'max_norm': max_norm, 'name': name}

        helper = LayerHelper("dgc_clip_by_norm_op", **args)

        if name is None:
1157 1158
            name = unique_name.generate_with_ignorable_key(".".join(
                [helper.name, 'tmp']))
1159 1160 1161 1162 1163

        out = helper.create_variable(
            type=x.type, name=name, dtype=x.dtype, persistable=False)

        helper.append_op(
G
gongweibao 已提交
1164
            type="dgc_clip_by_norm",
1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176
            inputs={"X": x,
                    "current_step": self._global_step_var},
            attrs={
                "max_norm": max_norm,
                "rampup_begin_step": float(self._rampup_begin_step)
            },
            outputs={"Out": out})
        return out

    def _append_clip_norm(self, grad_var, clip_norm):
        with grad_var.block.program._backward_role_guard():
            return self._clip_by_norm(
G
gongweibao 已提交
1177
                x=grad_var, max_norm=clip_norm, name=grad_var.name)
1178 1179

    def _dgc_op(self, param_var, clip_var, grad_var, u_var, v_var, k_var,
1180
                encoded_var, gather_var):
1181 1182
        block = framework.default_main_program().global_block()
        op_maker = core.op_proto_and_checker_maker
1183

1184 1185 1186 1187 1188 1189
        dgc_op = block.append_op(
            type="dgc",
            inputs={
                "U": u_var,
                "V": v_var,
                "Grad": clip_var,
1190
                "Param": param_var,
1191 1192
                "current_step": self._global_step_var,
                "nranks": self._nranks_var,
1193 1194 1195 1196 1197 1198
            },
            outputs={
                "U_out": u_var,
                "V_out": v_var,
                "EncodeGrad": encoded_var,
                "k": k_var,
1199 1200
                "Grad_out": grad_var,
                "GatherBuff": gather_var,
1201 1202 1203 1204 1205 1206
            },
            attrs={
                "m": self._momentum,
                "sparsity": self._sparsity,
                "use_nesterov": self._use_nesterov,
                "rampup_begin_step": float(self._rampup_begin_step),
1207
                "rampup_step": float(self._rampup_step),
1208 1209
                "regular_coeff": float(self.regular_coeff),
                "regular_type": int(self.regular_type),
1210 1211 1212 1213 1214 1215 1216 1217
            },
            stop_gradient=True)

        backward = op_maker.OpRole.Backward
        dgc_op._set_attr(op_maker.kOpRoleAttrName(), backward)
        dgc_op._set_attr(op_maker.kOpRoleVarAttrName(),
                         [param_var.name, grad_var.name])

1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248
    def apply_gradients(self, params_grads):
        params_grads = sorted(params_grads, key=lambda x: x[0].name)

        params_grads, table_param_and_grad, table_optimize_op = \
            self._process_distribute_lookuptable(params_grads)

        not_dgc_params_grads = []
        dgc_params_grads = []
        for param, grad in params_grads:
            if not self._is_use_dgc(param, grad):
                not_dgc_params_grads.append((param, grad))
            else:
                dgc_params_grads.append((param, grad))

        # DGC clip and regularization in local
        not_dgc_params_grads = append_gradient_clip_ops(not_dgc_params_grads)

        # Add regularization if any
        not_dgc_params_grads = append_regularization_ops(not_dgc_params_grads,
                                                         self.regularization)

        params_grads = not_dgc_params_grads + dgc_params_grads
        params_grads = sorted(params_grads, key=lambda x: x[0].name)

        optimize_ops = self._create_optimization_pass(params_grads)
        if table_optimize_op is not None:
            optimize_ops.append(table_optimize_op)
            params_grads.append(table_param_and_grad)

        return optimize_ops

1249

1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264
class LarsMomentumOptimizer(Optimizer):
    """
    Momentum optimizer with LARS support

    The update equations are as follows:

    .. math::

        & local\_learning\_rate = learning\_rate * lars\_coeff * \\
          \\frac{||param||}{||gradient|| + lars\_weight\_decay * ||param||}

        & velocity = mu * velocity + local\_learning\_rate * (gradient + lars\_weight\_decay * param)

        & param = param - velocity

1265 1266 1267 1268 1269 1270 1271 1272 1273 1274
    Parameters:
        learning_rate (float|Variable): The learning rate used to update parameters. \
            Can be a float value or a Variable with one float value as data element. \
            momentum (float): momentum factor
        lars_coeff (float): Defines how much we trust the layer to change its weights.
        lars_weight_decay (float): Weight decay coefficient for decaying using LARS.
        regularization: A Regularizer, such as :ref:`api_fluid_regularizer_L2DecayRegularizer`.
            Optional, default is None.
        name (str, optional): This parameter is used by developers to print debugging information. \
            For details, please refer to :ref:`api_guide_Name`. Default is None.
1275 1276 1277 1278

    Examples:
        .. code-block:: python

1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294
            import paddle.fluid as fluid
            import numpy as np

            np_inp = np.array([[1.0, 2.0], [3.0, 4.0]], dtype=np.float32)
            inp = fluid.layers.data(
                name="inp", shape=[2, 2], append_batch_size=False)
            out = fluid.layers.fc(inp, size=3)
            out = fluid.layers.reduce_sum(out)
            optimizer = fluid.optimizer.LarsMomentumOptimizer(learning_rate=0.001, momentum=0.9)
            optimizer.minimize(out)

            exe = fluid.Executor(fluid.CPUPlace())
            exe.run(fluid.default_startup_program())
            exe.run(
                feed={"inp": np_inp},
                fetch_list=[out.name])
1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343
    """
    _velocity_acc_str = "velocity"

    def __init__(self,
                 learning_rate,
                 momentum,
                 lars_coeff=0.001,
                 lars_weight_decay=0.0005,
                 regularization=None,
                 name=None):
        assert learning_rate is not None
        assert momentum is not None
        super(LarsMomentumOptimizer, self).__init__(
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
        self.type = "lars_momentum"
        self._momentum = momentum
        self._lars_coeff = float(lars_coeff)
        self._lars_weight_decay = float(lars_weight_decay)

    def _create_accumulators(self, block, parameters):
        assert isinstance(block, framework.Block)

        for p in parameters:
            self._add_accumulator(self._velocity_acc_str, p)

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)

        velocity_acc = self._get_accumulator(self._velocity_acc_str,
                                             param_and_grad[0])
        # create the momentum optimize op
        momentum_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "Velocity": velocity_acc,
                "LearningRate": self._create_param_lr(param_and_grad)
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "VelocityOut": velocity_acc
            },
            attrs={
                "mu": self._momentum,
                "lars_coeff": self._lars_coeff,
                "lars_weight_decay": self._lars_weight_decay
M
minqiyang 已提交
1344 1345
            },
            stop_gradient=True)
1346 1347 1348 1349

        return momentum_op


1350
class AdagradOptimizer(Optimizer):
Q
qiaolongfei 已提交
1351
    """
1352 1353
    The Adaptive Gradient optimizer (Adagrad for short) can adaptively assign
    different learning rates to individual parameters.
Q
qiaolongfei 已提交
1354

1355
    The parameter ``param_out`` update rule with gradient ``grad``:
Q
qiaolongfei 已提交
1356 1357 1358 1359 1360 1361 1362

    .. math::

        moment\_out &= moment + grad * grad

        param\_out &= param - \\frac{learning\_rate * grad}{\sqrt{moment\_out} + \epsilon}

1363 1364 1365 1366 1367 1368
    Related paper: `Adaptive Subgradient Methods for Online Learning and
    Stochastic Optimization <http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf>`_.

    The original paper does not have the ``epsilon`` attribute. It is added here
    in our implementation as also proposed `Per-parameter adaptive learning rate
    methods <http://cs231n.github.io/neural-networks-3/#ada>`_
Q
qiaolongfei 已提交
1369 1370 1371
    for numerical stability to avoid the division by zero error.

    Args:
1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382
        learning_rate (float|Variable): The learning rate used to update ``Parameter``.
            It can be a float value or a ``Variable`` with a float type.
        epsilon (float, optional): A small float value for numerical stability.
            The default value is 1e-06.
        regularization (WeightDecayRegularizer, optional): A ``Regularizer``, such as
             :ref:`api_fluid_regularizer_L2DecayRegularizer`. The default value is None.
        name (str, optional): Normally there is no need for user to set this property.
            For more information, please refer to :ref:`api_guide_Name`.
            The default value is None.
        initial_accumulator_value (float, optional): Initial value for moment accumulator.
            The default value is 0.0.
Q
qiaolongfei 已提交
1383 1384 1385 1386

    Examples:
        .. code-block:: python

1387
            import numpy as np
1388
            import paddle.fluid as fluid
1389 1390

            np_inp = np.array([[1.0, 2.0], [3.0, 4.0]], dtype=np.float32)
1391
            inp = fluid.data(name="inp", shape=[2, 2])
1392 1393
            out = fluid.layers.fc(inp, size=3)
            out = fluid.layers.reduce_sum(out)
1394
            optimizer = fluid.optimizer.AdagradOptimizer(learning_rate=0.2)
1395 1396 1397 1398 1399 1400 1401
            optimizer.minimize(out)

            exe = fluid.Executor(fluid.CPUPlace())
            exe.run(fluid.default_startup_program())
            exe.run(
                feed={"inp": np_inp},
                fetch_list=[out.name])
1402 1403 1404
    """
    _moment_acc_str = "moment"

X
Xin Pan 已提交
1405 1406 1407 1408
    def __init__(self,
                 learning_rate,
                 epsilon=1.0e-6,
                 regularization=None,
1409
                 name=None,
X
xuezhong 已提交
1410
                 initial_accumulator_value=0.0):
1411 1412
        assert learning_rate is not None
        assert epsilon is not None
Q
Qiao Longfei 已提交
1413
        super(AdagradOptimizer, self).__init__(
X
Xin Pan 已提交
1414 1415 1416
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
1417 1418
        self.type = "adagrad"
        self._epsilon = epsilon
1419
        self.initial_accumulator_value = initial_accumulator_value
1420 1421 1422 1423 1424

    def _create_accumulators(self, block, parameters):
        assert isinstance(block, framework.Block)

        for p in parameters:
Q
Qiao Longfei 已提交
1425
            self._add_accumulator(self._moment_acc_str, p)
1426 1427 1428 1429 1430 1431

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)

        moment_acc = self._get_accumulator(self._moment_acc_str,
                                           param_and_grad[0])
1432 1433 1434 1435 1436 1437 1438 1439 1440 1441
        startup_block = framework.default_startup_program().global_block()
        startup_block.append_op(
            type='fill_constant',
            inputs={},
            outputs={'Out': [moment_acc]},
            attrs={
                'dtype': moment_acc.dtype,
                'value': self.initial_accumulator_value,
                'shape': moment_acc.shape,
            })
1442

1443
        # Create the adagrad optimizer op
1444 1445 1446 1447 1448 1449
        adagrad_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "Moment": moment_acc,
1450
                "LearningRate": self._create_param_lr(param_and_grad)
1451 1452 1453
            },
            outputs={"ParamOut": param_and_grad[0],
                     "MomentOut": moment_acc},
M
minqiyang 已提交
1454 1455
            attrs={"epsilon": self._epsilon},
            stop_gradient=True)
1456 1457

        return adagrad_op
1458 1459 1460


class AdamOptimizer(Optimizer):
Q
qiaolongfei 已提交
1461
    """
1462 1463 1464 1465 1466 1467
    The Adam optimzier uses an optimization described at the end
    of section 2 of `Adam paper <https://arxiv.org/abs/1412.6980>`_ ,
    it can dynamically adjusts the learning rate of each parameter using
    the 1st moment estimates and the 2nd moment estimates of the gradient.
    
    The parameter ``param_out`` update rule with gradient ``grad``:
Q
qiaolongfei 已提交
1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481

    .. math::

        t & = t + 1

        moment\_1\_out & = {\\beta}_1 * moment\_1 + (1 - {\\beta}_1) * grad

        moment\_2\_out & = {\\beta}_2 * moment\_2 + (1 - {\\beta}_2) * grad * grad

        learning\_rate & = learning\_rate * \\
                          \\frac{\sqrt{1 - {\\beta}_2^t}}{1 - {\\beta}_1^t}

        param\_out & = param - learning\_rate * \\frac{moment\_1}{\sqrt{moment\_2} + \epsilon}

1482 1483
    Related paper: `Adam: A Method for Stochastic Optimization <https://arxiv.org/abs/1412.6980>`_

Q
qiaolongfei 已提交
1484
    Args:
1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504
        learning_rate (float|Variable, optional): The learning rate used to update ``Parameter``.
            It can be a float value or a ``Variable`` with a float type. The default value is 0.001.
        beta1 (float, optional): The exponential decay rate for the 1st moment estimates.
            The default value is 0.9.
        beta2 (float, optional): The exponential decay rate for the 2nd moment estimates.
            The default value is 0.999.
        epsilon (float, optional): A small float value for numerical stability.
            The default value is 1e-08.
        regularization (WeightDecayRegularizer, optional): A ``Regularizer``, such as
             :ref:`api_fluid_regularizer_L2DecayRegularizer`. The default value is None.
        name (str, optional): Normally there is no need for user to set this property.
            For more information, please refer to :ref:`api_guide_Name`.
            The default value is None.
        lazy_mode (bool, optional): The official Adam algorithm has two moving-average accumulators.
            The accumulators are updated at every step. Every element of the two moving-average
            is updated in both dense mode and sparse mode. If the size of parameter is very large,
            then the update may be very slow. The lazy mode only update the element that has
            gradient in current mini-batch, so it will be much more faster. But this mode has
            different semantics with the original Adam algorithm and may lead to different result.
            The default value is False.
Q
qiaolongfei 已提交
1505 1506 1507 1508

    Examples:
        .. code-block:: python

1509 1510 1511 1512 1513 1514
            import paddle
            import paddle.fluid as fluid

            place = fluid.CPUPlace()
            main = fluid.Program()
            with fluid.program_guard(main):
1515 1516
                x = fluid.data(name='x', shape=[None, 13], dtype='float32')
                y = fluid.data(name='y', shape=[None, 1], dtype='float32')
1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531
                y_predict = fluid.layers.fc(input=x, size=1, act=None)
                cost = fluid.layers.square_error_cost(input=y_predict, label=y)
                avg_cost = fluid.layers.mean(cost)

                adam_optimizer = fluid.optimizer.AdamOptimizer(0.01)
                adam_optimizer.minimize(avg_cost)

                fetch_list = [avg_cost]
                train_reader = paddle.batch(
                    paddle.dataset.uci_housing.train(), batch_size=1)
                feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
                exe = fluid.Executor(place)
                exe.run(fluid.default_startup_program())
                for data in train_reader():
                    exe.run(main, feed=feeder.feed(data), fetch_list=fetch_list)
Q
qiaolongfei 已提交
1532

1533 1534 1535
    """
    _moment1_acc_str = "moment1"
    _moment2_acc_str = "moment2"
Q
qiaolongfei 已提交
1536 1537
    _beta1_pow_acc_str = "beta1_pow_acc"
    _beta2_pow_acc_str = "beta2_pow_acc"
1538 1539 1540 1541 1542

    def __init__(self,
                 learning_rate=0.001,
                 beta1=0.9,
                 beta2=0.999,
1543
                 epsilon=1e-8,
X
Xin Pan 已提交
1544
                 regularization=None,
Q
Qiao Longfei 已提交
1545
                 name=None,
Q
Qiao Longfei 已提交
1546
                 lazy_mode=False):
1547 1548 1549 1550
        assert learning_rate is not None
        assert beta1 is not None
        assert beta2 is not None
        assert epsilon is not None
Q
Qiao Longfei 已提交
1551
        super(AdamOptimizer, self).__init__(
X
Xin Pan 已提交
1552 1553 1554
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
1555 1556 1557 1558
        self.type = "adam"
        self._beta1 = beta1
        self._beta2 = beta2
        self._epsilon = epsilon
Q
Qiao Longfei 已提交
1559
        self._lazy_mode = lazy_mode
1560 1561 1562 1563 1564 1565

    def _create_accumulators(self, block, parameters):
        assert isinstance(block, framework.Block)

        # Create accumulator tensors for first and second moments
        for p in parameters:
Q
Qiao Longfei 已提交
1566 1567
            self._add_accumulator(self._moment1_acc_str, p)
            self._add_accumulator(self._moment2_acc_str, p)
Q
qiaolongfei 已提交
1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579
            self._add_accumulator(
                name=self._beta1_pow_acc_str,
                param=p,
                dtype='float32',
                fill_value=self._beta1,
                shape=[1])
            self._add_accumulator(
                name=self._beta2_pow_acc_str,
                param=p,
                dtype='float32',
                fill_value=self._beta2,
                shape=[1])
1580 1581 1582 1583 1584 1585 1586 1587

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)

        moment1 = self._get_accumulator(self._moment1_acc_str,
                                        param_and_grad[0])
        moment2 = self._get_accumulator(self._moment2_acc_str,
                                        param_and_grad[0])
Q
qiaolongfei 已提交
1588 1589 1590 1591 1592
        beta1_pow_acc = self._get_accumulator(self._beta1_pow_acc_str,
                                              param_and_grad[0])
        beta2_pow_acc = self._get_accumulator(self._beta2_pow_acc_str,
                                              param_and_grad[0])

1593
        # create the adam optimize op
1594 1595 1596 1597 1598
        adam_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
1599
                "LearningRate": self._create_param_lr(param_and_grad),
1600 1601
                "Moment1": moment1,
                "Moment2": moment2,
Q
qiaolongfei 已提交
1602 1603
                "Beta1Pow": beta1_pow_acc,
                "Beta2Pow": beta2_pow_acc
1604 1605 1606 1607 1608 1609 1610 1611 1612
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "Moment1Out": moment1,
                "Moment2Out": moment2
            },
            attrs={
                "beta1": self._beta1,
                "beta2": self._beta2,
Q
Qiao Longfei 已提交
1613
                "epsilon": self._epsilon,
1614 1615
                "lazy_mode": self._lazy_mode,
                "min_row_size_to_use_multithread": 1000
M
minqiyang 已提交
1616 1617
            },
            stop_gradient=True)
1618 1619 1620

        return adam_op

1621
    def _finish_update(self, block, param_and_grads):
1622 1623 1624
        """Update Beta1 and Beta2 Power accumulators
        """
        assert isinstance(block, framework.Block)
Q
Qiao Longfei 已提交
1625
        main_block = block.program.global_block()
1626
        for param, grad in param_and_grads:
C
chengduo 已提交
1627
            if grad is None or param.trainable is False:
1628
                continue
X
Xin Pan 已提交
1629 1630
            with param.block.program._optimized_guard(
                [param, grad]), name_scope("optimizer"):
1631 1632 1633 1634 1635 1636 1637 1638
                beta1_pow_acc = self._get_accumulator(self._beta1_pow_acc_str,
                                                      param)
                beta2_pow_acc = self._get_accumulator(self._beta2_pow_acc_str,
                                                      param)
                main_block.append_op(
                    type="scale",
                    inputs={"X": beta1_pow_acc},
                    outputs={"Out": beta1_pow_acc},
M
minqiyang 已提交
1639 1640
                    attrs={"scale": self._beta1},
                    stop_gradient=True)
1641 1642 1643 1644 1645

                main_block.append_op(
                    type="scale",
                    inputs={"X": beta2_pow_acc},
                    outputs={"Out": beta2_pow_acc},
M
minqiyang 已提交
1646 1647
                    attrs={"scale": self._beta2},
                    stop_gradient=True)
1648 1649 1650


class AdamaxOptimizer(Optimizer):
Q
qiaolongfei 已提交
1651
    """
1652 1653 1654 1655
    The Adamax optimizer is implemented based on the Adamax Optimization 
    in Section 7 of `Adam paper <https://arxiv.org/abs/1412.6980>`_.
    The Adamax algorithm is a variant of the Adam algorithm based on the infinite norm,
    which makes the learning rate update algorithm more stable and simple.
Q
qiaolongfei 已提交
1656

1657
    The parameter ``param_out`` update rule with gradient ``grad``:
Q
qiaolongfei 已提交
1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670

    .. math::

        t & = t + 1

        moment\_out & = {\\beta}_1 * moment + (1 - {\\beta}_1) * grad

        inf\_norm\_out & = max({\\beta}_2 * inf\_norm + \epsilon, |grad|)

        learning\_rate & = \\frac{learning\_rate}{1 - {\\beta}_1^t}

        param\_out & = param - learning\_rate * \\frac{moment\_out}{inf\_norm\_out}

1671
    Related paper: `Adam: A Method for Stochastic Optimization <https://arxiv.org/abs/1412.6980>`_
Q
qiaolongfei 已提交
1672

1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692
    The original paper does not have an ``epsilon`` attribute,
    it is added here for numerical stability to prevent the division by 0 error.

    Args:
        learning_rate (float|Variable, optional): The learning rate used to update ``Parameter``.
            It can be a float value or a ``Variable`` with a float type. The default value is 0.001.
        beta1 (float, optional): The exponential decay rate for the 1st moment estimates.
            The default value is 0.9.
        beta2 (float, optional): The exponential decay rate for the 2nd moment estimates.
            The default value is 0.999.
        epsilon (float, optional): A small float value for numerical stability.
            The default value is 1e-08.
        regularization (WeightDecayRegularizer, optional): A ``Regularizer``, such as
             :ref:`api_fluid_regularizer_L2DecayRegularizer`. The default value is None.
        name (str, optional): Normally there is no need for user to set this property.
            For more information, please refer to :ref:`api_guide_Name`.
            The default value is None.

    **Notes**:
        **Currently, AdamaxOptimizer doesn't support sparse parameter optimization.**
Q
qiaolongfei 已提交
1693

1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706
    Examples:
        .. code-block:: python

          import paddle.fluid as fluid
          import numpy

          # First create the Executor.
          place = fluid.CPUPlace() # fluid.CUDAPlace(0)
          exe = fluid.Executor(place)

          train_program = fluid.Program()
          startup_program = fluid.Program()
          with fluid.program_guard(train_program, startup_program):
1707
              data = fluid.data(name='X', shape=[None, 1], dtype='float32')
1708 1709
              hidden = fluid.layers.fc(input=data, size=10)
              loss = fluid.layers.mean(hidden)
1710
              adam = fluid.optimizer.AdamaxOptimizer(learning_rate=0.2)
1711 1712 1713 1714 1715 1716 1717 1718 1719
              adam.minimize(loss)

          # Run the startup program once and only once.
          exe.run(startup_program)

          x = numpy.random.random(size=(10, 1)).astype('float32')
          outs = exe.run(program=train_program,
                        feed={'X': x},
                         fetch_list=[loss.name])
1720 1721 1722
    """
    _moment_acc_str = "moment"
    _inf_norm_acc_str = "inf_norm"
Q
qiaolongfei 已提交
1723
    _beta1_pow_acc_str = "beta1_pow_acc"
1724 1725 1726 1727 1728

    def __init__(self,
                 learning_rate=0.001,
                 beta1=0.9,
                 beta2=0.999,
1729
                 epsilon=1e-8,
X
Xin Pan 已提交
1730 1731
                 regularization=None,
                 name=None):
1732 1733 1734 1735
        assert learning_rate is not None
        assert beta1 is not None
        assert beta2 is not None
        assert epsilon is not None
Q
Qiao Longfei 已提交
1736
        super(AdamaxOptimizer, self).__init__(
X
Xin Pan 已提交
1737 1738 1739
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
1740 1741 1742 1743 1744 1745 1746 1747
        self.type = "adamax"
        self._beta1 = beta1
        self._beta2 = beta2
        self._epsilon = epsilon

    def _create_accumulators(self, block, parameters):
        # Create accumulator tensors for first moment and infinity norm
        for p in parameters:
Q
Qiao Longfei 已提交
1748 1749
            self._add_accumulator(self._moment_acc_str, p)
            self._add_accumulator(self._inf_norm_acc_str, p)
Q
qiaolongfei 已提交
1750 1751 1752 1753 1754 1755
            self._add_accumulator(
                name=self._beta1_pow_acc_str,
                param=p,
                dtype='float32',
                fill_value=self._beta1,
                shape=[1])
1756 1757 1758 1759 1760 1761 1762

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)

        moment = self._get_accumulator(self._moment_acc_str, param_and_grad[0])
        inf_norm = self._get_accumulator(self._inf_norm_acc_str,
                                         param_and_grad[0])
Q
qiaolongfei 已提交
1763 1764
        beta1_pow_acc = self._get_accumulator(self._beta1_pow_acc_str,
                                              param_and_grad[0])
1765 1766 1767 1768 1769 1770
        # create the adamax optimize op
        adamax_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
1771
                "LearningRate": self._create_param_lr(param_and_grad),
1772 1773
                "Moment": moment,
                "InfNorm": inf_norm,
Q
qiaolongfei 已提交
1774
                "Beta1Pow": beta1_pow_acc
1775 1776 1777 1778 1779 1780 1781 1782 1783 1784
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "MomentOut": moment,
                "InfNormOut": inf_norm
            },
            attrs={
                "beta1": self._beta1,
                "beta2": self._beta2,
                "epsilon": self._epsilon
M
minqiyang 已提交
1785 1786
            },
            stop_gradient=True)
1787 1788 1789

        return adamax_op

1790
    def _finish_update(self, block, parameters_and_grads):
1791 1792 1793
        """Update Beta1 Power accumulator
        """
        assert isinstance(block, framework.Block)
Q
Qiao Longfei 已提交
1794
        main_block = block.program.global_block()
1795
        for param, grad in parameters_and_grads:
C
chengduo 已提交
1796
            if grad is None or param.trainable is False:
1797
                continue
X
Xin Pan 已提交
1798 1799
            with param.block.program._optimized_guard(
                [param, grad]), name_scope('adamx'):
1800 1801 1802 1803 1804 1805
                beta1_pow_acc = self._get_accumulator(self._beta1_pow_acc_str,
                                                      param)
                main_block.append_op(
                    type="scale",
                    inputs={"X": beta1_pow_acc},
                    outputs={"Out": beta1_pow_acc},
M
minqiyang 已提交
1806 1807
                    attrs={"scale": self._beta1},
                    stop_gradient=True)
1808 1809


1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888
class DpsgdOptimizer(Optimizer):
    """
    We implement the Dpsgd optimizer according to CCS16 paper -
    Deep Learning with Differential Privacy.

    Examples:
        .. code-block:: python

          import paddle.fluid as fluid
          import numpy

          # First create the Executor.
          place = fluid.CPUPlace() # fluid.CUDAPlace(0)
          exe = fluid.Executor(place)

          train_program = fluid.Program()
          startup_program = fluid.Program()
          with fluid.program_guard(train_program, startup_program):
              data = fluid.layers.data(name='X', shape=[1], dtype='float32')
              hidden = fluid.layers.fc(input=data, size=10)
              loss = fluid.layers.mean(hidden)
              optimizer = fluid.optimizer.Dpsgd(learning_rate=0.01, clip=10.0, batch_size=16.0, sigma=1.0)
              optimizer.minimize(loss)

          # Run the startup program once and only once.
          exe.run(startup_program)

          x = numpy.random.random(size=(10, 1)).astype('float32')
          outs = exe.run(program=train_program,
                        feed={'X': x},
                         fetch_list=[loss.name])

    Args:
        learning_rate (float|Variable): the learning rate used to update parameters. \
        Can be a float value or a Variable with one float value as data element.
        clip (float): clipping threshold
        batch_size (float): batch size.
        sigma (float): for gaussian noise.
    Notes:
       Currently, DpsgdOptimizer doesn't support sparse parameter optimization.
    """

    def __init__(self,
                 learning_rate=0.001,
                 clip=0.9,
                 batch_size=0.999,
                 sigma=1e-8):
        assert learning_rate is not None
        assert clip is not None
        assert batch_size is not None
        assert sigma is not None
        super(DpsgdOptimizer, self).__init__(learning_rate=learning_rate)
        self.type = "dpsgd"
        self._clip = clip
        self._batch_size = batch_size
        self._sigma = sigma

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)

        # create the dpsgd optimize op
        dpsgd_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "LearningRate": self._create_param_lr(param_and_grad)
            },
            outputs={"ParamOut": param_and_grad[0]},
            attrs={
                "clip": self._clip,
                "batch_size": self._batch_size,
                "sigma": self._sigma
            },
            stop_gradient=True)

        return dpsgd_op


1889
class DecayedAdagradOptimizer(Optimizer):
1890
    """
1891 1892 1893
    The Decayed Adagrad optimizer can be seen as an Adagrad algorithm that introduces
    the decay rate to solve the problem of a sharp drop in the learning rate
    during model training when using the AdagradOptimizer.
1894

1895
    The parameter ``param_out`` update rule with gradient ``grad``:
1896 1897 1898 1899 1900 1901 1902

    .. math::

        moment\_out & = decay * moment + (1 - decay) * grad * grad

        param\_out & = param - \\frac{learning\_rate * grad}{\sqrt{moment\_out} + \epsilon}

1903 1904 1905 1906
    Related paper: `Adaptive Subgradient Methods for Online Learning and Stochastic
    Optimization <http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf>`_.

    The original paper does not have an ``epsilon`` attribute. It is added here for numerical
1907 1908 1909
    stability to avoid the division by zero error.

    Args:
1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922
        learning_rate (float|Variable): The learning rate used to update ``Parameter``.
            It can be a float value or a ``Variable`` with a float type.
        decay (float, optional): The decay rate. The default value is 0.95.
        epsilon (float, optional): A small float value for numerical stability.
            The default value is 1e-06.
        regularization (WeightDecayRegularizer, optional): A ``Regularizer``, such as
             :ref:`api_fluid_regularizer_L2DecayRegularizer`. The default value is None.
        name (str, optional): Normally there is no need for user to set this property.
            For more information, please refer to :ref:`api_guide_Name`.
            The default value is None.

    **Notes**:
        **Currently, DecayedAdagradOptimizer doesn't support sparse parameter optimization.**
1923 1924 1925 1926

    Examples:
        .. code-block:: python

1927 1928
            import paddle.fluid as fluid

1929 1930 1931 1932
            x = fluid.data( name='x', shape=[None, 10], dtype='float32' )
            trans = fluid.layers.fc( x, 100 )
            cost = fluid.layers.reduce_mean( trans )
            optimizer = fluid.optimizer.DecayedAdagradOptimizer(learning_rate=0.2)
1933
            optimizer.minimize(cost)
1934 1935 1936
    """
    _moment_acc_str = "moment"

X
Xin Pan 已提交
1937 1938 1939 1940 1941 1942
    def __init__(self,
                 learning_rate,
                 decay=0.95,
                 epsilon=1.0e-6,
                 regularization=None,
                 name=None):
1943 1944 1945 1946
        assert learning_rate is not None
        assert decay is not None
        assert epsilon is not None

Q
Qiao Longfei 已提交
1947
        super(DecayedAdagradOptimizer, self).__init__(
X
Xin Pan 已提交
1948 1949 1950
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977
        self.type = "decayed_adagrad"
        self._decay = decay
        self._epsilon = epsilon

    def _create_accumulators(self, block, parameters):
        assert isinstance(block, framework.Block)

        for p in parameters:
            self._add_accumulator(self._moment_acc_str, p)

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)

        moment_acc = self._get_accumulator(self._moment_acc_str,
                                           param_and_grad[0])

        # Create the decayed adagrad optimizer op
        decayed_adagrad_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "Moment": moment_acc,
                "LearningRate": self._create_param_lr(param_and_grad)
            },
            outputs={"ParamOut": param_and_grad[0],
                     "MomentOut": moment_acc},
M
minqiyang 已提交
1978 1979
            attrs={"epsilon": self._epsilon},
            stop_gradient=True)
1980 1981

        return decayed_adagrad_op
1982 1983


1984
class AdadeltaOptimizer(Optimizer):
1985
    """
Z
Zeng Jinle 已提交
1986
    **Notes: This API does not support sparse parameter optimization.**
Q
qiaolongfei 已提交
1987

Z
Zeng Jinle 已提交
1988
    Adadelta Optimizer. Please refer to this for details:
Z
Zeng Jinle 已提交
1989 1990 1991
    `ADADELTA: AN ADAPTIVE LEARNING RATE METHOD <https://arxiv.org/abs/1212.5701>`_.

    The update is done as follows:
1992

Z
Zeng Jinle 已提交
1993 1994
    .. math::

Z
Zeng Jinle 已提交
1995
        E(g_t^2) &= \\rho * E(g_{t-1}^2) + (1-\\rho) * g^2
1996

Z
Zeng Jinle 已提交
1997
        learning\_rate &= \sqrt{ ( E(dx_{t-1}^2) + \\epsilon ) / ( E(g_t^2) + \\epsilon ) }
Z
Zeng Jinle 已提交
1998

Z
Zeng Jinle 已提交
1999
        E(dx_t^2) &= \\rho * E(dx_{t-1}^2) + (1-\\rho) * (-g*learning\_rate)^2
2000 2001

    Args:
Z
Zeng Jinle 已提交
2002 2003 2004 2005 2006 2007
        learning_rate (float|Variable): global learning rate.
        epsilon (float): a small float number for numeric stability. Default 1.0e-6.
        rho (float): a floating point value indicating the decay rate. Default 0.95.
        regularization (WeightDecayRegularizer, optional): A Regularizer, such as
                fluid.regularizer.L2DecayRegularizer. Default None, meaning that there is no
                regularization.
2008 2009 2010
        name (str, optional): The default value is None. Normally there is no need for user
                to set this property. For more information, please refer to
                :ref:`api_guide_Name` .
2011 2012 2013 2014

    Examples:
        .. code-block:: python

2015
            import paddle.fluid as fluid
Z
Zeng Jinle 已提交
2016

2017
            image = fluid.data(name='image', shape=[None, 28], dtype='float32')
Z
Zeng Jinle 已提交
2018 2019
            fc = fluid.layers.fc(image, size=10)
            cost = fluid.layers.reduce_mean(fc)
2020 2021
            optimizer = fluid.optimizer.Adadelta(
                learning_rate=0.0003, epsilon=1.0e-6, rho=0.95)
C
chengduo 已提交
2022

Z
Zeng Jinle 已提交
2023 2024 2025 2026
            # optimizer_ops is a list of optimizer operators to update parameters
            # params_grads is a list of (param, param_grad), where param is each
            # parameter and param_grad is the gradient variable of param.
            optimizer_ops, params_grads = optimizer.minimize(cost)
2027
    """
2028

2029 2030 2031
    _avg_squared_grad_acc_str = "_avg_squared_grad"
    _avg_squared_update_acc_str = "_avg_squared_update"

X
Xin Pan 已提交
2032 2033 2034 2035 2036 2037
    def __init__(self,
                 learning_rate,
                 epsilon=1.0e-6,
                 rho=0.95,
                 regularization=None,
                 name=None):
2038 2039 2040 2041 2042 2043
        if learning_rate is None:
            raise ValueError("learning_rate is not set.")
        if epsilon is None:
            raise ValueError("epsilon is not set.")
        if rho is None:
            raise ValueError("rho is not set.")
2044
        super(AdadeltaOptimizer, self).__init__(
X
Xin Pan 已提交
2045 2046 2047
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
2048 2049 2050 2051 2052
        self.type = "adadelta"
        self._epsilon = epsilon
        self._rho = rho

    def _create_accumulators(self, block, parameters):
2053 2054
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")
2055 2056 2057 2058 2059 2060

        for p in parameters:
            self._add_accumulator(self._avg_squared_grad_acc_str, p)
            self._add_accumulator(self._avg_squared_update_acc_str, p)

    def _append_optimize_op(self, block, param_and_grad):
2061 2062
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")
2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083

        avg_squared_grad_acc = self._get_accumulator(
            self._avg_squared_grad_acc_str, param_and_grad[0])
        avg_squared_update_acc = self._get_accumulator(
            self._avg_squared_update_acc_str, param_and_grad[0])

        # Create the adadelta optimizer op
        adadelta_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "AvgSquaredGrad": avg_squared_grad_acc,
                "AvgSquaredUpdate": avg_squared_update_acc
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "AvgSquaredGradOut": avg_squared_grad_acc,
                "AvgSquaredUpdateOut": avg_squared_update_acc
            },
            attrs={"epsilon": self._epsilon,
M
minqiyang 已提交
2084 2085
                   "rho": self._rho},
            stop_gradient=True)
2086 2087 2088 2089

        return adadelta_op


Q
qingqing01 已提交
2090 2091 2092 2093 2094 2095 2096 2097 2098 2099
class RMSPropOptimizer(Optimizer):
    """
    Root Mean Squared Propagation (RMSProp) is an unpublished, adaptive learning
    rate method. The original slides proposed RMSProp: Slide 29 of
    http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf .

    The original equation is as follows:

    ..  math::

Q
qiaolongfei 已提交
2100
        r(w, t) & = \\rho r(w, t-1) + (1 - \\rho)(\\nabla Q_{i}(w))^2
Q
qingqing01 已提交
2101 2102 2103 2104

        w & = w - \\frac{\\eta} {\\sqrt{r(w,t) + \\epsilon}} \\nabla Q_{i}(w)

    The first equation calculates moving average of the squared gradient for
Q
qiaolongfei 已提交
2105
    each weight. Then dividing the gradient by :math:`sqrt{v(w,t)}`.
Q
qingqing01 已提交
2106 2107 2108 2109 2110 2111

    In some cases, adding a momentum term :math: `\\beta` is beneficial.
    In our implementation, Nesterov momentum is used:

    ..  math::

Q
qiaolongfei 已提交
2112
        r(w, t) & = \\rho r(w, t-1) + (1 - \\rho)(\\nabla Q_{i}(w))^2
Q
qingqing01 已提交
2113

2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127
        v(w, t) & = \\beta v(w, t-1) + \\frac{\\eta} {\\sqrt{r(w,t) +
            \\epsilon}} \\nabla Q_{i}(w)

        w & = w - v(w, t)

    if centered is True:

    ..  math::

        r(w, t) & = \\rho r(w, t-1) + (1 - \\rho)(\\nabla Q_{i}(w))^2

        g(w, t) & = \\rho g(w, t-1) + (1 - \\rho)\\nabla Q_{i}(w)

        v(w, t) & = \\beta v(w, t-1) + \\frac{\\eta} {\\sqrt{r(w,t) - (g(w, t))^2 +
Q
qingqing01 已提交
2128 2129 2130 2131
            \\epsilon}} \\nabla Q_{i}(w)

        w & = w - v(w, t)

Q
qiaolongfei 已提交
2132
    where, :math:`\\rho` is a hyperparameter and typical values are 0.9, 0.95
Q
qingqing01 已提交
2133 2134 2135 2136 2137
    and so on. :math: `beta` is the momentum term. :math: `\\epsilon` is a
    smoothing term to avoid division by zero, usually set somewhere in range
    from 1e-4 to 1e-8.


2138 2139 2140
    Parameters:
        learning_rate(float): Global learning rate.
        rho(float): rho is :math: `\\rho` in equation, default is 0.95.
Q
qingqing01 已提交
2141
        epsilon(float): :math: `\\epsilon` in equation is smoothing term to
2142
            avoid division by zero, default is 1e-6.
Q
qiaolongfei 已提交
2143
        momentum(float): :math:`\\beta` in equation is the momentum term,
2144
            default is 0.0.
2145 2146 2147 2148
        centered(bool): If True, gradients are normalized by the estimated variance of
            the gradient; if False, by the uncentered second moment. Setting this to
            True may help with training, but is slightly more expensive in terms of
            computation and memory. Defaults to False.
2149 2150 2151 2152
        regularization: A Regularizer, such as :ref:`api_fluid_regularizer_L2DecayRegularizer`. \
            Optional, default is None.
        name (str, optional): This parameter is used by developers to print debugging information. \
            For details, please refer to :ref:`api_guide_Name`. Default is None.
Q
qingqing01 已提交
2153 2154 2155 2156 2157 2158 2159

    Raises:
        ValueError: If learning_rate, rho, epsilon, momentum are None.

    Examples:
          .. code-block:: python

2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184
            import paddle
            import paddle.fluid as fluid
            import numpy as np

            place = fluid.CPUPlace()
            main = fluid.Program()
            with fluid.program_guard(main):
                x = fluid.layers.data(name='x', shape=[13], dtype='float32')
                y = fluid.layers.data(name='y', shape=[1], dtype='float32')
                y_predict = fluid.layers.fc(input=x, size=1, act=None)
                cost = fluid.layers.square_error_cost(input=y_predict, label=y)
                avg_cost = fluid.layers.mean(cost)

                rms_optimizer = fluid.optimizer.RMSProp(learning_rate=0.1)
                rms_optimizer.minimize(avg_cost)

                fetch_list = [avg_cost]
                train_reader = paddle.batch(
                    paddle.dataset.uci_housing.train(), batch_size=1)
                feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
                exe = fluid.Executor(place)
                exe.run(fluid.default_startup_program())
                for data in train_reader():
                    exe.run(main, feed=feeder.feed(data), fetch_list=fetch_list)

Q
qingqing01 已提交
2185 2186 2187 2188
    """

    _momentum_acc_str = "momentum"
    _mean_square_acc_str = "mean_square"
2189
    _mean_grad_acc_str = "mean_grad"
Q
qingqing01 已提交
2190 2191 2192 2193 2194 2195

    def __init__(self,
                 learning_rate,
                 rho=0.95,
                 epsilon=1.0e-6,
                 momentum=0.0,
2196
                 centered=False,
X
Xin Pan 已提交
2197 2198
                 regularization=None,
                 name=None):
Q
qingqing01 已提交
2199
        super(RMSPropOptimizer, self).__init__(
X
Xin Pan 已提交
2200 2201 2202
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
Q
qingqing01 已提交
2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215
        if learning_rate is None:
            raise ValueError("learning_rate is not set.")
        if rho is None:
            raise ValueError("rho is not set.")
        if epsilon is None:
            raise ValueError("epsilon is not set.")
        if momentum is None:
            raise ValueError("momentum is not set.")

        self.type = "rmsprop"
        self._rho = rho
        self._epsilon = epsilon
        self._momentum = momentum
2216
        self._centered = centered
Q
qingqing01 已提交
2217 2218 2219 2220 2221 2222 2223 2224

    def _create_accumulators(self, block, parameters):
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")

        for p in parameters:
            self._add_accumulator(self._momentum_acc_str, p)
            self._add_accumulator(self._mean_square_acc_str, p)
2225
            self._add_accumulator(self._mean_grad_acc_str, p)
Q
qingqing01 已提交
2226 2227 2228 2229 2230 2231 2232 2233 2234

    def _append_optimize_op(self, block, param_and_grad):
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")

        momentum_acc = self._get_accumulator(self._momentum_acc_str,
                                             param_and_grad[0])
        mean_square_acc = self._get_accumulator(self._mean_square_acc_str,
                                                param_and_grad[0])
2235 2236
        mean_grad_acc = self._get_accumulator(self._mean_grad_acc_str,
                                              param_and_grad[0])
Q
qingqing01 已提交
2237 2238 2239 2240 2241 2242 2243
        rmsprop_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "Moment": momentum_acc,
                "MeanSquare": mean_square_acc,
2244
                "MeanGrad": mean_grad_acc,
Q
qingqing01 已提交
2245 2246 2247 2248 2249
                "LearningRate": self._create_param_lr(param_and_grad),
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "MomentOut": momentum_acc,
2250 2251
                "MeanSquareOut": mean_square_acc,
                "MeanGradOut": mean_grad_acc
Q
qingqing01 已提交
2252 2253 2254 2255
            },
            attrs={
                "epsilon": self._epsilon,
                "decay": self._rho,
2256 2257
                "momentum": self._momentum,
                "centered": self._centered
M
minqiyang 已提交
2258 2259
            },
            stop_gradient=True)
Q
qingqing01 已提交
2260 2261 2262 2263

        return rmsprop_op


Q
qiaolongfei 已提交
2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303
class FtrlOptimizer(Optimizer):
    """
    FTRL (Follow The Regularized Leader) Optimizer.

    The paper that proposed Follow The Regularized Leader (FTRL):
    (https://www.eecs.tufts.edu/~dsculley/papers/ad-click-prediction.pdf)

    ..  math::

        &new\_accum = squared\_accum + grad^2

        &if (lr\_power == -0.5):

        &\quad  linear\_accum += grad - \\frac{\\sqrt{new\_accum} - \\sqrt{squared\_accum}}{learning\_rate * param}

        &else:

        &\quad   linear\_accum += grad - \\frac{new\_accum^{-lr\_power} - accum^{-lr\_power}}{learning\_rate * param}


        &x = l1 * sign(linear\_accum) - linear\_accum

        &if (lr\_power == -0.5):

        &\quad   y = \\frac{\\sqrt{new\_accum}}{learning\_rate} + (2 * l2)

        &\quad   pre\_shrink = \\frac{x}{y}

        &\quad   param = (abs(linear\_accum) > l1).select(pre\_shrink, 0.0)

        &else:

        &\quad   y = \\frac{new\_accum^{-lr\_power}}{learning\_rate} + (2 * l2)

        &\quad   pre\_shrink = \\frac{x}{y}

        &\quad   param = (abs(linear\_accum) > l1).select(pre\_shrink, 0.0)

        &squared\_accum += grad^2

2304 2305 2306 2307 2308 2309 2310 2311 2312
    Parameters:
        learning_rate (float|Variable): Global learning rate.
        l1 (float): L1 regularization strength, default is 0.0.
        l2 (float): L2 regularization strength, default is 0.0.
        lr_power (float): Learning Rate Power, default is -0.5.
        regularization: A Regularizer, such as :ref:`api_fluid_regularizer_L2DecayRegularizer`. \
            Optional, default is None.
        name (str, optional): This parameter is used by developers to print debugging information. \
            For details, please refer to :ref:`api_guide_Name`. Default is None.
Q
qiaolongfei 已提交
2313 2314 2315 2316 2317 2318 2319

    Raises:
        ValueError: If learning_rate, rho, epsilon, momentum are None.

    Examples:
          .. code-block:: python

2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343
            import paddle
            import paddle.fluid as fluid
            import numpy as np

            place = fluid.CPUPlace()
            main = fluid.Program()
            with fluid.program_guard(main):
                x = fluid.layers.data(name='x', shape=[13], dtype='float32')
                y = fluid.layers.data(name='y', shape=[1], dtype='float32')
                y_predict = fluid.layers.fc(input=x, size=1, act=None)
                cost = fluid.layers.square_error_cost(input=y_predict, label=y)
                avg_cost = fluid.layers.mean(cost)

                ftrl_optimizer = fluid.optimizer.Ftrl(learning_rate=0.1)
                ftrl_optimizer.minimize(avg_cost)

                fetch_list = [avg_cost]
                train_reader = paddle.batch(
                    paddle.dataset.uci_housing.train(), batch_size=1)
                feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
                exe = fluid.Executor(place)
                exe.run(fluid.default_startup_program())
                for data in train_reader():
                    exe.run(main, feed=feeder.feed(data), fetch_list=fetch_list)
C
chengduo 已提交
2344

2345
    NOTE:
C
chengduo 已提交
2346
       Currently, FtrlOptimizer doesn't support sparse parameter optimization.
Q
qiaolongfei 已提交
2347 2348 2349 2350 2351
    """

    _squared_acc_str = "squared"
    _linear_acc_str = "linear"

X
Xin Pan 已提交
2352 2353 2354 2355 2356 2357 2358
    def __init__(self,
                 learning_rate,
                 l1=0.0,
                 l2=0.0,
                 lr_power=-0.5,
                 regularization=None,
                 name=None):
Q
qiaolongfei 已提交
2359
        super(FtrlOptimizer, self).__init__(
X
Xin Pan 已提交
2360 2361 2362
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
Q
qiaolongfei 已提交
2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402
        if learning_rate is None:
            raise ValueError("learning_rate is not set.")

        self.type = "ftrl"
        self._l1 = l1
        self._l2 = l2
        self._lr_power = lr_power

    def _create_accumulators(self, block, parameters):
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")

        for p in parameters:
            self._add_accumulator(self._squared_acc_str, p)
            self._add_accumulator(self._linear_acc_str, p)

    def _append_optimize_op(self, block, param_and_grad):
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")

        squared_acc = self._get_accumulator(self._squared_acc_str,
                                            param_and_grad[0])
        linear_acc = self._get_accumulator(self._linear_acc_str,
                                           param_and_grad[0])
        ftrl_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "SquaredAccumulator": squared_acc,
                "LinearAccumulator": linear_acc,
                "LearningRate": self._create_param_lr(param_and_grad),
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "SquaredAccumOut": squared_acc,
                "LinearAccumOut": linear_acc
            },
            attrs={"l1": self._l1,
                   "l2": self._l1,
M
minqiyang 已提交
2403 2404
                   "lr_power": self._lr_power},
            stop_gradient=True)
Q
qiaolongfei 已提交
2405 2406 2407 2408

        return ftrl_op


Y
Yibing Liu 已提交
2409 2410 2411 2412 2413 2414
class LambOptimizer(AdamOptimizer):
    """
    LAMB (Layer-wise Adaptive Moments optimizer for Batching training) Optimizer.

    LAMB Optimizer is designed to scale up the batch size of training without losing 
    accuracy, which supports adaptive element-wise updating and accurate layer-wise 
Y
Yibing Liu 已提交
2415 2416
    correction. For more information, please refer to `Large Batch Optimization for 
    Deep Learning: Training BERT in 76 minutes <https://arxiv.org/abs/1904.00962>`_ .
Y
Yibing Liu 已提交
2417 2418 2419 2420 2421

    The updating of parameters follows:

    ..  math::

Y
Yibing Liu 已提交
2422
        m_t &= \\beta_1 m_{t - 1}+ (1 - \\beta_1)g_t 
Y
Yibing Liu 已提交
2423

Y
Yibing Liu 已提交
2424
        v_t &= \\beta_2 v_{t - 1}  + (1 - \\beta_2)g_t^2
Y
Yibing Liu 已提交
2425

Y
Yibing Liu 已提交
2426
        r_t &= \\frac{m_t}{\\sqrt{v_t}+\\epsilon}
Y
Yibing Liu 已提交
2427

Y
Yibing Liu 已提交
2428
        w_t &= w_{t-1} -\\eta_t \\frac{\\left \| w_{t-1}\\right \|}{\\left \| r_t + \\lambda w_{t-1}\\right \|} (r_t + \\lambda w_{t-1})
Y
Yibing Liu 已提交
2429 2430 2431 2432 2433 2434


    where :math:`m` is the 1st moment, and :math:`v` the 2nd moment, :math:`\\eta` the 
    learning rate, :math:`\\lambda` the LAMB weight decay rate.

    Args:
Y
Yibing Liu 已提交
2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449
        learning_rate (float|Variable, optional): the learning rate used to update parameters. \
            Can be a float value or a Variable with data type float32. Default 0.001.
        lamb_weight_decay (float, optional): The LAMB weight decay rate. Default 0.01.
        beta1 (float, optional): The exponential decay rate for the 1st moment estimates.
            Default 0.9.
        beta2 (float, optional): The exponential decay rate for the 2nd moment estimates.
            Default 0.999.
        epsilon (float, optional): A small float value for numerical stability. Default 1e-6.
        regularization (Regularizer|None): A Regularizer, such as
           fluid.regularizer.L1DecayRegularizer. Default None.
        exclude_from_weight_decay_fn (function|None): Exclude a parameter from weight 
            decay when **exclude_from_weight_decay_fn(parameter)** returns true. 
            Default None.
        name(str|None): For detailed information, please refer to 
            :ref:`api_guide_Name` . Usually name is no need to set and None by default.
Y
Yibing Liu 已提交
2450 2451 2452 2453 2454 2455

    Examples:
        .. code-block:: python
            
            import paddle.fluid as fluid 

Y
Yibing Liu 已提交
2456
            data = fluid.data(name='x', shape=[-1, 5], dtype='float32')
Y
Yibing Liu 已提交
2457 2458 2459
            hidden = fluid.layers.fc(input=data, size=10)
            cost = fluid.layers.mean(hidden)

Y
Yibing Liu 已提交
2460 2461 2462 2463 2464
            def exclude_fn(param):
                return param.name.endswith('.b_0')

            optimizer = fluid.optimizer.Lamb(learning_rate=0.002,
                                             exclude_from_weight_decay_fn=exclude_fn)
Y
Yibing Liu 已提交
2465 2466 2467 2468
            optimizer.minimize(cost)
    """
    _moment1_acc_str = "moment1"
    _moment2_acc_str = "moment2"
Y
Yibing Liu 已提交
2469
    # these two not used in op temporarily
Y
Yibing Liu 已提交
2470 2471 2472 2473 2474 2475 2476 2477 2478 2479
    _beta1_pow_acc_str = "beta1_pow_acc"
    _beta2_pow_acc_str = "beta2_pow_acc"

    def __init__(self,
                 learning_rate=0.001,
                 lamb_weight_decay=0.01,
                 beta1=0.9,
                 beta2=0.999,
                 epsilon=1e-6,
                 regularization=None,
Y
Yibing Liu 已提交
2480
                 exclude_from_weight_decay_fn=None,
Y
Yibing Liu 已提交
2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495
                 name=None):
        assert learning_rate is not None
        assert lamb_weight_decay is not None
        assert beta1 is not None
        assert beta2 is not None
        assert epsilon is not None
        super(LambOptimizer, self).__init__(
            learning_rate=learning_rate,
            regularization=regularization,
            beta1=beta1,
            beta2=beta2,
            epsilon=epsilon,
            name=name)
        self.type = "lamb"
        self._weight_decay = lamb_weight_decay
Y
Yibing Liu 已提交
2496
        self._exclude_from_weight_decay_fn = exclude_from_weight_decay_fn
Y
Yibing Liu 已提交
2497 2498 2499

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)
2500
        block.program._use_lamb = True
Y
Yibing Liu 已提交
2501 2502 2503 2504 2505 2506 2507 2508 2509 2510

        moment1 = self._get_accumulator(self._moment1_acc_str,
                                        param_and_grad[0])
        moment2 = self._get_accumulator(self._moment2_acc_str,
                                        param_and_grad[0])
        beta1_pow_acc = self._get_accumulator(self._beta1_pow_acc_str,
                                              param_and_grad[0])
        beta2_pow_acc = self._get_accumulator(self._beta2_pow_acc_str,
                                              param_and_grad[0])

Y
Yibing Liu 已提交
2511 2512 2513 2514 2515 2516
        if self._exclude_from_weight_decay_fn is not None \
            and self._exclude_from_weight_decay_fn(param_and_grad[0]):
            weight_decay = 0.0
        else:
            weight_decay = self._weight_decay

Y
Yibing Liu 已提交
2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537
        # create the lamb optimize op
        lamb_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "LearningRate": self._create_param_lr(param_and_grad),
                "Moment1": moment1,
                "Moment2": moment2,
                "Beta1Pow": beta1_pow_acc,
                "Beta2Pow": beta2_pow_acc
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "Moment1Out": moment1,
                "Moment2Out": moment2
            },
            attrs={
                "beta1": self._beta1,
                "beta2": self._beta2,
                "epsilon": self._epsilon,
Y
Yibing Liu 已提交
2538
                "weight_decay": weight_decay
Y
Yibing Liu 已提交
2539 2540 2541 2542 2543 2544
            },
            stop_gradient=True)

        return lamb_op


2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557
# We short the class name, since users will use the optimizer with the package
# name. The sample code:
#
# import paddle.fluid as fluid
#
# sgd = fluid.optimizer.SGD(...)
#
# It is no need to add an `Optimizer` as the class suffix
SGD = SGDOptimizer
Momentum = MomentumOptimizer
Adagrad = AdagradOptimizer
Adam = AdamOptimizer
Adamax = AdamaxOptimizer
2558
Dpsgd = DpsgdOptimizer
2559
DecayedAdagrad = DecayedAdagradOptimizer
2560
Adadelta = AdadeltaOptimizer
Q
qingqing01 已提交
2561
RMSProp = RMSPropOptimizer
Q
qiaolongfei 已提交
2562
Ftrl = FtrlOptimizer
2563
LarsMomentum = LarsMomentumOptimizer
Y
Yibing Liu 已提交
2564
Lamb = LambOptimizer
2565 2566 2567


class ModelAverage(Optimizer):
2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586
    """
    The ModelAverage optimizer accumulates specific continuous historical parameters
    during training. The accumulated historical range can be controlled by the passed
    ``average_window_rate`` argument. The averaged ``Parameter`` are used in the prediction,
    which usually can improve the accuracy of the prediction.

    Accumulate the average of the ``Parameter`` in the sliding window, the result will be saved
    in a temporary variable, can be applied to the current model's ``Parameter`` by calling
    the ``apply()`` method, and the current model ``Parameter`` can be restored by calling
    the ``restore()`` method.

    The window size for calculating the average is determined by ``average_window_rate``,
    ``min_average_window``, ``max_average_window`` and the current ``Parameter`` update times (num_updates).

    When the cumulative times (num_accumulates) is greater than the specific window
    threshold (average_window), the accumulated ``Parameter`` temporary variable is set to 0.0.
    The following example will help to understand the role of these arguments:

    ::
2587

2588 2589 2590 2591 2592 2593 2594 2595 2596
        if num_accumulates >= min_average_window and num_accumulates >= min(max_average_window, num_updates * average_window_rate):
            num_accumulates = 0

    In the above conditional judgment statement, ``num_accumulates`` indicates the current
    accumulated number, which can be abstractly understood as the length of the cumulative window.
    The length of the window must be at least the length set by the ``min_average_window`` argument,
    and cannot exceed the length specified by the ``max_average_window`` argument or
    ``num_updates * average_window_rate``, where ``num_updates`` indicates the current ``Parameter``
    update times, ``average_window_rate`` is a coefficient that calculates the length of the window.
2597 2598

    Args:
2599 2600 2601 2602 2603 2604 2605 2606
        average_window_rate (float): The calculate ratio of the window length relative to ``Parameter`` update times.
        min_average_window (int, optional): the minimum size of average window length. The default value is 10000.
        max_average_window (int, optional): The maximum size of average window length. The default value is 10000.
        regularization (WeightDecayRegularizer, optional): A ``Regularizer``, such as
             :ref:`api_fluid_regularizer_L2DecayRegularizer`. The default value is None.
        name (str, optional): Normally there is no need for user to set this property.
            For more information, please refer to :ref:`api_guide_Name`.
            The default value is None.
2607

2608
    Examples:
Q
qiaolongfei 已提交
2609 2610 2611

      .. code-block:: python

2612 2613 2614 2615 2616 2617
        import paddle.fluid as fluid
        import numpy

        # First create the Executor.
        place = fluid.CPUPlace()  # fluid.CUDAPlace(0)
        exe = fluid.Executor(place)
2618

2619 2620 2621 2622
        train_program = fluid.Program()
        startup_program = fluid.Program()
        with fluid.program_guard(train_program, startup_program):
            # build net
2623
            data = fluid.data(name='X', shape=[None, 1], dtype='float32')
2624 2625 2626 2627 2628 2629 2630 2631
            hidden = fluid.layers.fc(input=data, size=10)
            loss = fluid.layers.mean(hidden)
            optimizer = fluid.optimizer.Momentum(learning_rate=0.2, momentum=0.1)
            optimizer.minimize(loss)

            # build ModelAverage optimizer
            model_average = fluid.optimizer.ModelAverage(0.15,
                                                         min_average_window=10000,
2632
                                                         max_average_window=12500)
2633 2634

            exe.run(startup_program)
2635 2636 2637 2638 2639
            for i in range(12500):
                x = numpy.random.random(size=(10, 1)).astype('float32')
                outs = exe.run(program=train_program,
                               feed={'X': x},
                               fetch_list=[loss.name])
2640 2641

            # apply ModelAverage
2642
            with model_average.apply(exe):
2643 2644 2645 2646
                x = numpy.random.random(size=(10, 1)).astype('float32')
                exe.run(program=train_program,
                        feed={'X': x},
                        fetch_list=[loss.name])
2647 2648 2649
    """

    def __init__(self,
W
wanghaoshuang 已提交
2650
                 average_window_rate,
2651 2652
                 min_average_window=10000,
                 max_average_window=10000,
X
Xin Pan 已提交
2653 2654 2655 2656
                 regularization=None,
                 name=None):
        super(ModelAverage, self).__init__(
            0.0, regularization=regularization, name=name)
2657 2658 2659
        self.average_window = average_window_rate
        self.min_average_window = min_average_window
        self.max_average_window = max_average_window
2660

2661
        self.params_grads = []
2662 2663
        for param in framework.default_main_program().global_block(
        ).all_parameters():
2664
            if param.do_model_average != False:
2665
                grad = param.block.create_var(
2666 2667
                    name=unique_name.generate_with_ignorable_key(".".join(
                        [param.name, 'tmp'])),
2668 2669
                    dtype=param.dtype,
                    persistable=False,
W
wanghaoshuang 已提交
2670
                    stop_gradient=True)
2671
                self.params_grads.append((param, grad))
2672

2673
        for param, grad in self.params_grads:
2674 2675
            if grad is None:
                continue
X
Xin Pan 已提交
2676 2677
            with param.block.program._optimized_guard(
                [param, grad]), name_scope('move_average'):
2678
                self._append_average_accumulate_op(param)
2679

2680 2681 2682 2683
        self.apply_program = Program()
        block = self.apply_program.global_block()
        with program_guard(main_program=self.apply_program):
            for param_grad in self.params_grads:
2684
                self._add_average_apply_op(block, param_grad)
2685 2686 2687 2688 2689

        self.restore_program = Program()
        block = self.restore_program.global_block()
        with program_guard(main_program=self.restore_program):
            for param_grad in self.params_grads:
2690
                self._add_average_restore_op(block, param_grad)
2691

2692
    def _add_average_apply_op(self, block, param_grad):
L
Luo Tao 已提交
2693 2694 2695 2696 2697 2698
        param = block._clone_variable(param_grad[0])
        grad = block._clone_variable(param_grad[1])
        sum_1 = block._clone_variable(self._get_accumulator('sum_1', param))
        sum_2 = block._clone_variable(self._get_accumulator('sum_2', param))
        sum_3 = block._clone_variable(self._get_accumulator('sum_3', param))
        num_accumulates = block._clone_variable(
2699
            self._get_accumulator('num_accumulates', param))
L
Luo Tao 已提交
2700
        old_num_accumulates = block._clone_variable(
2701
            self._get_accumulator('old_num_accumulates', param))
L
Luo Tao 已提交
2702
        num_updates = block._clone_variable(
2703 2704 2705 2706 2707 2708
            self._get_accumulator('num_updates', param))
        # backup param value to grad
        layers.assign(input=param, output=grad)
        # param = (sum_1 + sum_2 + sum_3) / (num_accumulates + old_num_accumulates)
        tmp = layers.sum(x=[num_accumulates, old_num_accumulates])
        sum = layers.sum(x=[sum_1, sum_2, sum_3])
D
dzhwinter 已提交
2709 2710 2711 2712
        tmp = layers.cast(
            x=tmp, dtype='float32' if self._dtype == None else self._dtype)
        sum = layers.cast(
            x=sum, dtype='float32' if self._dtype == None else self._dtype)
S
sneaxiy 已提交
2713
        ops._elementwise_div(x=sum, y=tmp, out=param)
2714 2715

    def _add_average_restore_op(self, block, param_grad):
L
Luo Tao 已提交
2716 2717
        param = block._clone_variable(param_grad[0])
        grad = block._clone_variable(param_grad[1])
2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754
        layers.assign(input=grad, output=param)

    def _append_average_accumulate_op(self, param):
        self.helper = LayerHelper("average_accumulate")
        sum_1 = self._add_accumulator('sum_1', param)
        sum_2 = self._add_accumulator('sum_2', param)
        sum_3 = self._add_accumulator('sum_3', param)
        num_accumulates = self._add_accumulator(
            'num_accumulates', param, dtype='int64', shape=[1])
        old_num_accumulates = self._add_accumulator(
            'old_num_accumulates', param, dtype='int64', shape=[1])
        num_updates = self._add_accumulator(
            'num_updates', param, dtype='int64', shape=[1])

        self.helper.append_op(
            type='average_accumulates',
            inputs={
                "param": param,
                "in_sum_1": sum_1,
                "in_sum_2": sum_2,
                "in_sum_3": sum_3,
                "in_num_accumulates": num_accumulates,
                "in_old_num_accumulates": old_num_accumulates,
                "in_num_updates": num_updates
            },
            outputs={
                "out_sum_1": sum_1,
                "out_sum_2": sum_2,
                "out_sum_3": sum_3,
                "out_num_accumulates": num_accumulates,
                "out_old_num_accumulates": old_num_accumulates,
                "out_num_updates": num_updates,
            },
            attrs={
                "average_window": self.average_window,
                "min_average_window": self.min_average_window,
                "max_average_window": self.max_average_window,
M
minqiyang 已提交
2755 2756
            },
            stop_gradient=True)
2757

S
rename  
sneaxiy 已提交
2758
    @signature_safe_contextmanager
2759
    def apply(self, executor, need_restore=True):
2760 2761
        """
        Apply the average of the cumulative ``Parameter`` to the parameters of the current model.
2762 2763

        Args:
2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807
            executor(fluid.Executor): The current network executor.
            need_restore(bool): Restore flag variable, if set to True, the network will restore
                the parameters of the network to the default value, if set to False,
                it will not be restored. The default value is True.

        Examples:

          .. code-block:: python

            import paddle.fluid as fluid
            import numpy

            # First create the Executor.
            place = fluid.CPUPlace()  # fluid.CUDAPlace(0)
            exe = fluid.Executor(place)

            train_program = fluid.Program()
            startup_program = fluid.Program()
            with fluid.program_guard(train_program, startup_program):
                # build net
                data = fluid.data(name='X', shape=[None, 1], dtype='float32')
                hidden = fluid.layers.fc(input=data, size=10)
                loss = fluid.layers.mean(hidden)
                optimizer = fluid.optimizer.Momentum(learning_rate=0.2, momentum=0.1)
                optimizer.minimize(loss)

                # build ModelAverage optimizer
                model_average = fluid.optimizer.ModelAverage(0.15,
                                                            min_average_window=10000,
                                                            max_average_window=12500)

                exe.run(startup_program)
                for i in range(12500):
                    x = numpy.random.random(size=(10, 1)).astype('float32')
                    outs = exe.run(program=train_program,
                                feed={'X': x},
                                fetch_list=[loss.name])

                # apply ModelAverage
                with model_average.apply(exe):
                    x = numpy.random.random(size=(10, 1)).astype('float32')
                    exe.run(program=train_program,
                            feed={'X': x},
                            fetch_list=[loss.name])
2808
        """
2809 2810 2811 2812 2813 2814
        executor.run(self.apply_program)
        try:
            yield
        finally:
            if need_restore:
                self.restore(executor)
2815 2816

    def restore(self, executor):
2817 2818
        """
        Restore ``Parameter`` values of current model.
2819 2820
        
        Args:
2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864
            executor(fluid.Executor): The current network executor.

        Examples:

          .. code-block:: python

            import paddle.fluid as fluid
            import numpy

            # First create the Executor.
            place = fluid.CPUPlace()  # fluid.CUDAPlace(0)
            exe = fluid.Executor(place)

            train_program = fluid.Program()
            startup_program = fluid.Program()
            with fluid.program_guard(train_program, startup_program):
                # build net
                data = fluid.data(name='X', shape=[None, 1], dtype='float32')
                hidden = fluid.layers.fc(input=data, size=10)
                loss = fluid.layers.mean(hidden)
                optimizer = fluid.optimizer.Momentum(learning_rate=0.2, momentum=0.1)
                optimizer.minimize(loss)

                # build ModelAverage optimizer
                model_average = fluid.optimizer.ModelAverage(0.15,
                                                            min_average_window=10000,
                                                            max_average_window=12500)

                exe.run(startup_program)
                for i in range(12500):
                    x = numpy.random.random(size=(10, 1)).astype('float32')
                    outs = exe.run(program=train_program,
                                feed={'X': x},
                                fetch_list=[loss.name])

                # apply ModelAverage
                with model_average.apply(exe, False):
                    x = numpy.random.random(size=(10, 1)).astype('float32')
                    exe.run(program=train_program,
                            feed={'X': x},
                            fetch_list=[loss.name])

                # restore Parameters
                model_average.restore(exe)
2865
        """
2866
        executor.run(self.restore_program)
2867 2868 2869 2870 2871 2872 2873 2874 2875 2876


class ExponentialMovingAverage(object):
    """
    Compute the moving average of parameters with exponential decay.
    Given a parameter :math:`\\theta`, its exponential moving average (EMA)
    will be

    ..  math::

2877
        \\text{EMA}_0 & = 0
2878

2879 2880
	\\text{EMA}_t & = \\text{decay} * \\text{EMA}_{t-1} + (1 - \\text{decay}) * \\theta_t

Y
Yibing Liu 已提交
2881 2882 2883 2884
    The average results calculated by **update()** method will be saved in 
    temporary variables which are created and maintained by the object, and can 
    be applied to parameters of current model by calling **apply()** method. And 
    the **restore()** method is used to restore the parameters.
2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905

    **Bias correction**. All EMAs are initialized to :math:`0` and hence they will be 
    zero biased, which can be corrected by divided by a factor 
    :math:`(1 - \\text{decay}^t)` , i.e., the actual EMAs applied to parameters 
    when calling **apply()** method would be 

    ..  math::
    
        \\widehat{\\text{EMA}}_t = \\frac{\\text{EMA}_t}{1 - \\text{decay}^t}

    **Decay rate scheduling**. A large decay rate very close to 1 would result 
    in that the averages move very slowly. And a better strategy is to set a 
    relative smaller decay rate in the very beginning. The argument **thres_steps**
    allows users to pass a Variable to schedule the decay rate, in this case, 
    the actual decay rate becomes
     
    ..  math::
    
        \\min(\\text{decay}, \\frac{1 + \\text{thres_steps}}{10 + \\text{thres_steps}})

    Usually **thres_steps** can be the global training steps.
2906 2907 2908


    Args:
Y
Yibing Liu 已提交
2909 2910 2911 2912 2913 2914 2915
	decay (float, optional): The exponential decay rate, usually close to 1, such as 
            0.999, 0.9999, ... . Default 0.999.
        thres_steps (Variable|None): If not `None`, schedule the decay rate. 
            Default None.
        name (str|None): For detailed information, please refer to 
            :ref:`api_guide_Name`. Usually name is no need to set and None by 
            default.
2916 2917 2918 2919 2920


    Examples:

	.. code-block:: python
2921 2922 2923 2924 2925

	    import numpy
	    import paddle
	    import paddle.fluid as fluid

Y
Yibing Liu 已提交
2926
	    data = fluid.data(name='x', shape=[-1, 5], dtype='float32')
2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964
	    hidden = fluid.layers.fc(input=data, size=10)
	    cost = fluid.layers.mean(hidden)

	    test_program = fluid.default_main_program().clone(for_test=True)

	    optimizer = fluid.optimizer.Adam(learning_rate=0.001)
	    optimizer.minimize(cost)

	    global_steps = fluid.layers.learning_rate_scheduler._decay_step_counter()
	    ema = fluid.optimizer.ExponentialMovingAverage(0.999, thres_steps=global_steps)
	    ema.update()

	    place = fluid.CPUPlace()
	    exe = fluid.Executor(place)
	    exe.run(fluid.default_startup_program())

	    for pass_id in range(3):
		for batch_id in range(6):
		    data = numpy.random.random(size=(10, 5)).astype('float32')
		    exe.run(program=fluid.default_main_program(),
			feed={'x': data}, 
			fetch_list=[cost.name])

		# usage 1
		with ema.apply(exe):
		    data = numpy.random.random(size=(10, 5)).astype('float32')
		    exe.run(program=test_program,
			    feed={'x': data}, 
			    fetch_list=[hidden.name])
			    

		 # usage 2
		with ema.apply(exe, need_restore=False):
		    data = numpy.random.random(size=(10, 5)).astype('float32')
		    exe.run(program=test_program,
			    feed={'x': data}, 
			    fetch_list=[hidden.name])
		ema.restore(exe)
2965 2966
    """

2967
    def __init__(self, decay=0.999, thres_steps=None, name=None):
2968
        self._decay = decay
2969
        self._thres_steps = thres_steps
2970
        self._name = name if name is not None else ''
2971 2972
        self._decay_var = self._get_ema_decay()

Y
Yibing Liu 已提交
2973
        self._params_tmps = []
2974
        for param in default_main_program().global_block().all_parameters():
2975 2976 2977 2978 2979 2980 2981
            if param.do_model_average != False:
                tmp = param.block.create_var(
                    name=unique_name.generate(".".join(
                        [self._name + param.name, 'ema_tmp'])),
                    dtype=param.dtype,
                    persistable=False,
                    stop_gradient=True)
Y
Yibing Liu 已提交
2982
                self._params_tmps.append((param, tmp))
2983

Y
Yibing Liu 已提交
2984 2985
        self._ema_vars = {}
        for param, tmp in self._params_tmps:
2986 2987
            with param.block.program._optimized_guard(
                [param, tmp]), name_scope('moving_average'):
Y
Yibing Liu 已提交
2988
                self._ema_vars[param.name] = self._create_ema_vars(param)
2989 2990 2991 2992

        self.apply_program = Program()
        block = self.apply_program.global_block()
        with program_guard(main_program=self.apply_program):
2993
            decay_pow = self._get_decay_pow(block)
Y
Yibing Liu 已提交
2994
            for param, tmp in self._params_tmps:
2995 2996
                param = block._clone_variable(param)
                tmp = block._clone_variable(tmp)
Y
Yibing Liu 已提交
2997
                ema = block._clone_variable(self._ema_vars[param.name])
2998
                layers.assign(input=param, output=tmp)
2999 3000
                # bias correction
                ema = ema / (1.0 - decay_pow)
3001 3002 3003 3004 3005
                layers.assign(input=ema, output=param)

        self.restore_program = Program()
        block = self.restore_program.global_block()
        with program_guard(main_program=self.restore_program):
Y
Yibing Liu 已提交
3006
            for param, tmp in self._params_tmps:
3007 3008 3009 3010
                tmp = block._clone_variable(tmp)
                param = block._clone_variable(param)
                layers.assign(input=tmp, output=param)

3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037
    def _get_ema_decay(self):
        with default_main_program()._lr_schedule_guard():
            decay_var = layers.tensor.create_global_var(
                shape=[1],
                value=self._decay,
                dtype='float32',
                persistable=True,
                name="scheduled_ema_decay_rate")

            if self._thres_steps is not None:
                decay_t = (self._thres_steps + 1.0) / (self._thres_steps + 10.0)
                with layers.control_flow.Switch() as switch:
                    with switch.case(decay_t < self._decay):
                        layers.tensor.assign(decay_t, decay_var)
                    with switch.default():
                        layers.tensor.assign(
                            np.array(
                                [self._decay], dtype=np.float32),
                            decay_var)
        return decay_var

    def _get_decay_pow(self, block):
        global_steps = layers.learning_rate_scheduler._decay_step_counter()
        decay_var = block._clone_variable(self._decay_var)
        decay_pow_acc = layers.elementwise_pow(decay_var, global_steps + 1)
        return decay_pow_acc

Y
Yibing Liu 已提交
3038
    def _create_ema_vars(self, param):
3039 3040 3041 3042 3043 3044 3045 3046 3047
        param_ema = layers.create_global_var(
            name=unique_name.generate(self._name + param.name + '_ema'),
            shape=param.shape,
            value=0.0,
            dtype=param.dtype,
            persistable=True)

        return param_ema

Y
Yibing Liu 已提交
3048 3049 3050 3051 3052
    def update(self):
        """ 
        Update Exponential Moving Average. Should only call this method in 
        train program.
        """
3053
        param_master_emas = []
Y
Yibing Liu 已提交
3054 3055 3056 3057
        for param, tmp in self._params_tmps:
            with param.block.program._optimized_guard(
                [param, tmp]), name_scope('moving_average'):
                param_ema = self._ema_vars[param.name]
3058
                if param.name + '.master' in self._ema_vars:
3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075
                    master_ema = self._ema_vars[param.name + '.master']
                    param_master_emas.append([param_ema, master_ema])
                else:
                    ema_t = param_ema * self._decay_var + param * (
                        1 - self._decay_var)
                    layers.assign(input=ema_t, output=param_ema)

        # for fp16 params
        for param_ema, master_ema in param_master_emas:
            default_main_program().global_block().append_op(
                type="cast",
                inputs={"X": master_ema},
                outputs={"Out": param_ema},
                attrs={
                    "in_dtype": master_ema.dtype,
                    "out_dtype": param_ema.dtype
                })
Y
Yibing Liu 已提交
3076

3077 3078 3079 3080 3081 3082 3083
    @signature_safe_contextmanager
    def apply(self, executor, need_restore=True):
        """
        Apply moving average to parameters for evaluation.
        
        Args:
            executor (Executor): The Executor to execute applying.
Y
Yibing Liu 已提交
3084 3085
            need_restore (bool, optional): Whether to restore parameters after 
                applying. Default True.
3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100
        """
        executor.run(self.apply_program)
        try:
            yield
        finally:
            if need_restore:
                self.restore(executor)

    def restore(self, executor):
        """Restore parameters.
        
        Args:
            executor (Executor): The Executor to execute restoring.
        """
        executor.run(self.restore_program)
H
hutuxian 已提交
3101 3102 3103


class PipelineOptimizer(object):
3104 3105
    """
    Pipeline Optimizer
H
hutuxian 已提交
3106 3107 3108 3109 3110 3111 3112 3113 3114

    Train with pipeline mode. The program will be splited by cut_list. 

    If the len of cut_list is k, then the whole program (including \
    backward part) will be splited to 2*k-1 sections. 
    
    So the length of place_list and concurrency_list must be also 2*k-1.

    Note: Though the asynchronous mode is applied in pipeline training to speed up, \
3115
    the final performance depends on the training progress of each pipeline heavily.
H
hutuxian 已提交
3116 3117 3118

    And we will try the synchronous mode in the future.

3119
    Args:
H
hutuxian 已提交
3120 3121 3122 3123
        optimizer (Optimizer): The based optimizer, such as SGD.
        cut_list (list of Variable list): The cut variable of the main_program.
        place_list (list of Place): The place where the section will run on.
        concurrency_list (list of int): The concurrency degree.
3124 3125
        queue_size (int): Each section will consume scopes from its in-scope queue 
                        and produce scopes to out-scope queue. And this parameter 
H
hutuxian 已提交
3126 3127 3128 3129
                        specify the scope queue size. [Optional. Default: 30].
        sync_steps (int): The synchronization steps between different cards. [Optional. Default: 1].
        start_cpu_core_id (int): specify the first cpu core id. [Optional. Default:0].

3130 3131
    Examples:
        .. code-block:: python
H
hutuxian 已提交
3132

3133
            import paddle.fluid as fluid
H
hutuxian 已提交
3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167
            import paddle.fluid.layers as layers

            x = fluid.layers.data(name='x', shape=[1], dtype='int64', lod_level=0)
            y = fluid.layers.data(name='y', shape=[1], dtype='int64', lod_level=0)
            emb_x = layers.embedding(input=x, param_attr=fluid.ParamAttr(name="embx"), size=[10,2], is_sparse=False)
            emb_y = layers.embedding(input=y, param_attr=fluid.ParamAttr(name="emby",learning_rate=0.9), size=[10,2], is_sparse=False)
            concat = layers.concat([emb_x, emb_y], axis=1)
            fc = layers.fc(input=concat, name="fc", size=1, num_flatten_dims=1, bias_attr=False)
            loss = layers.reduce_mean(fc)
            optimizer = fluid.optimizer.SGD(learning_rate=0.5)
            optimizer = fluid.optimizer.PipelineOptimizer(optimizer,
                    cut_list=[[emb_x, emb_y], [loss]],
                    place_list=[fluid.CPUPlace(), fluid.CUDAPlace(0), fluid.CPUPlace()],
                    concurrency_list=[1, 1, 4],
                    queue_size=2,
                    sync_steps=1,
                    )
            optimizer.minimize(loss)
            place = fluid.CPUPlace()
            exe = fluid.Executor(place)
            exe.run(fluid.default_startup_program())
            filelist = [] # you should set your own filelist, e.g. filelist = ["dataA.txt"]
            dataset = fluid.DatasetFactory().create_dataset("FileInstantDataset")
            dataset.set_use_var([x,y])
            dataset.set_batch_size(batch_size)
            dataset.set_filelist(filelist)
            exe.train_from_dataset(
                        fluid.default_main_program(),
                        dataset,
                        thread=2,
                        debug=False,
                        fetch_list=[],
                        fetch_info=[],
                        print_period=1)
3168 3169
    """

H
hutuxian 已提交
3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186
    def __init__(self,
                 optimizer,
                 cut_list=None,
                 place_list=None,
                 concurrency_list=None,
                 queue_size=30,
                 sync_steps=1,
                 start_cpu_core_id=0):
        # TODO: check properties
        self._optimizer = optimizer
        self._cut_list = cut_list
        self._place_list = place_list
        self._concurrency_list = concurrency_list
        self._queue_size = queue_size
        self._sync_steps = sync_steps
        self._start_cpu_core_id = start_cpu_core_id

H
hutuxian 已提交
3187
    def _create_vars(self, block, main_program):
H
hutuxian 已提交
3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198
        used_var_set = set()
        for op_idx in range(block.desc.op_size()):
            op_desc = block.desc.op(op_idx)
            vars = op_desc.input_arg_names() + op_desc.output_arg_names()
            for var in vars:
                if var in used_var_set:
                    continue
                used_var_set.add(var)
                source_var = main_program.block(0).var(str(var))
                block._clone_variable(source_var, False)

H
hutuxian 已提交
3199
    def _extract_section_opt_ops(self, ops, cut_point_name):
H
hutuxian 已提交
3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214
        """
        Extract opt ops in the given section
        """
        output_names = set(cut_point_name)
        relevant_op_flags = [True] * len(ops)
        for i, op in reversed(list(enumerate(ops))):
            if _some_in_set_(op.desc.output_arg_names(), output_names):
                for name in op.desc.input_arg_names():
                    output_names.add(name)
            else:
                relevant_op_flags[i] = False

        op_path = [ops[i] for i in range(len(ops)) if relevant_op_flags[i]]
        return op_path

H
hutuxian 已提交
3215
    def _find_input_output(self, ops, name, is_forward=True):
H
hutuxian 已提交
3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229
        """
        Find the inputs or outputs of a section
        """
        all_set = set()
        part_set = set()
        for op in ops:
            if is_forward:
                part_set.update(op.desc.output_arg_names())
            else:
                part_set.update(op.desc.input_arg_names())
            all_set.update(op.desc.output_arg_names())
            all_set.update(op.desc.input_arg_names())
        return all_set - part_set

H
hutuxian 已提交
3230
    def _find_persistable_vars(self, ops, whole_parameters):
H
hutuxian 已提交
3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257
        """
        find the persistable input vars in current section
        """
        res = set()
        for op in ops:
            vars = op.desc.input_arg_names()
            for var in vars:
                if var in whole_parameters:
                    res.add(var)
        return res

    def _is_opt_role_op(self, op):
        op_maker = core.op_proto_and_checker_maker
        optimize_role = core.op_proto_and_checker_maker.OpRole.Optimize
        if op_maker.kOpRoleAttrName() in op.attr_names and \
                int(op.all_attrs()[op_maker.kOpRoleAttrName()]) & int(optimize_role) != 0:
            return True
        return False

    def _is_lr_role_op(self, op):
        op_maker = core.op_proto_and_checker_maker
        optimize_role = core.op_proto_and_checker_maker.OpRole.LRSched
        if op_maker.kOpRoleAttrName() in op.attr_names and \
                int(op.all_attrs()[op_maker.kOpRoleAttrName()]) == int(optimize_role):
            return True
        return False

H
hutuxian 已提交
3258
    def _extract_section_ops(self, ops, cut_point_name):
H
hutuxian 已提交
3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277
        """
        Extract ops in the given section 
        """
        output_names = set(cut_point_name)
        relevant_op_flags = [True] * len(ops)
        for i, op in reversed(list(enumerate(ops))):
            if not self._is_opt_role_op(op) and _some_in_set_(
                    op.desc.output_arg_names(), output_names):
                for name in op.desc.input_arg_names():
                    output_names.add(name)
            elif op.desc.type() == "print" and op.desc.input_arg_names()[
                    0] in output_names:
                continue
            else:
                relevant_op_flags[i] = False

        op_path = [ops[i] for i in range(len(ops)) if relevant_op_flags[i]]
        return op_path

H
hutuxian 已提交
3278 3279
    def _find_section_opt(self, ops, params):
        res = self._extract_section_opt_ops(ops, params)
H
hutuxian 已提交
3280 3281
        return res

H
hutuxian 已提交
3282
    def _split_program(self, main_program, cut_list):
H
hutuxian 已提交
3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302
        programs = []
        block = main_program.block(0)
        whole_parameters = [e.name for e in block.all_parameters()]
        cut_var_names = []
        cut_len = len(cut_list)
        sec_params = []
        for i, cut_vars in enumerate(cut_list[:-1]):
            cut_var_names.append([cut_var.name for cut_var in cut_vars])
        for i, cut_vars in reversed(list(enumerate(cut_list[:-1]))):
            cut_var_names.append(
                [_append_grad_suffix_(cut_var.name) for cut_var in cut_vars])
            if i == 0:
                cut_var_names[-1] += [var.name for var in cut_list[-1]]
        ops = block.ops[:]
        for i, cut_vars in enumerate(cut_var_names):
            program = {
                "program": Program(),
                "input_set": set(),
                "output_set": set()
            }
H
hutuxian 已提交
3303
            cur_ops = self._extract_section_ops(ops, cut_vars)
H
hutuxian 已提交
3304 3305 3306 3307 3308 3309
            if i == 0:
                for op in ops:
                    if self._is_lr_role_op(op):
                        cur_ops.append(op)
            #prevent inplace in/out
            program["input_set"].update(
H
hutuxian 已提交
3310
                self._find_input_output(
H
hutuxian 已提交
3311 3312 3313 3314 3315 3316
                    cur_ops, [], is_forward=True))
            for e in cur_ops:
                ops.remove(e)

            if i < cut_len:
                sec_params.append(
H
hutuxian 已提交
3317
                    self._find_persistable_vars(cur_ops, whole_parameters))
H
hutuxian 已提交
3318
            if i >= cut_len - 1:
H
hutuxian 已提交
3319 3320
                opt_ops = self._find_section_opt(
                    ops, sec_params[2 * cut_len - 2 - i])
H
hutuxian 已提交
3321 3322 3323 3324 3325 3326 3327 3328 3329 3330

                for e in opt_ops:
                    ops.remove(e)
                cur_ops += opt_ops

            op_descs = [op.desc for op in cur_ops]
            for op_desc in op_descs:
                ap_op = program["program"].block(0).desc.append_op()
                ap_op.copy_from(op_desc)
            program["input_set"].update(
H
hutuxian 已提交
3331
                self._find_input_output(
H
hutuxian 已提交
3332 3333 3334
                    cur_ops, cut_vars, is_forward=True))
            program["input_set"].update(sec_params[min(i, 2 * cut_len - 2 - i)])
            program["output_set"].update(
H
hutuxian 已提交
3335
                self._find_input_output(
H
hutuxian 已提交
3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349
                    cur_ops, cut_vars, is_forward=False))
            programs.append(program)
        program = {
            "program": Program(),
            "input_set": set(),
            "output_set": set()
        }
        op_descs = [op.desc for op in ops]
        for op_desc in op_descs:
            ap_op = program["program"].block(0).desc.append_op()
            ap_op.copy_from(op_desc)
        program["input_set"].update(
            [cut_var.name + "@GRAD" for cut_var in cut_list[0]])
        program["input_set"].update(
H
hutuxian 已提交
3350
            self._find_input_output(
H
hutuxian 已提交
3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370
                ops, [], is_forward=True))
        program["input_set"].update(sec_params[0])
        programs.append(program)
        inputs = set()
        for program in reversed(list(programs)):
            output_list = list(program["output_set"])
            for output in output_list:
                if output not in inputs:
                    program["output_set"].remove(output)
            inputs.update(program["input_set"])
        return programs

    def minimize(self,
                 loss,
                 startup_program=None,
                 parameter_list=None,
                 no_grad_set=None):
        self._optimizer.minimize(loss, startup_program, parameter_list,
                                 no_grad_set)
        program = loss.block.program
H
hutuxian 已提交
3371
        program_list = self._split_program(program, self._cut_list)
H
hutuxian 已提交
3372
        for p in program_list:
H
hutuxian 已提交
3373
            self._create_vars(p["program"].block(0), program)
H
hutuxian 已提交
3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393
        whole_parameters = [e.name for e in program.block(0).all_parameters()]
        param_need_sync = []
        for i, section_p in enumerate(program_list):
            if not isinstance(self._place_list[i], core.CUDAPlace):
                continue
            section_var = [e for e in section_p["program"].block(0).vars]
            for p in section_var:
                if p in whole_parameters:
                    param_need_sync.append(p)
        program._pipeline_opt = {
            "trainer": "PipelineTrainer",
            "device_worker": "Section",
            "section_program_list": program_list,
            "place_list": self._place_list,
            "concurrency_list": self._concurrency_list,
            "queue_size": self._queue_size,
            "start_cpu_core_id": self._start_cpu_core_id,
            "sync_steps": self._sync_steps,
            "param_need_sync": param_need_sync
        }
M
mapingshuo 已提交
3394 3395


M
mapingshuo 已提交
3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623
class RecomputeOptimizer(Optimizer):
    """
    Recompute Optimizer Wrapper

    Normally, a training step contains three sub-steps: first, run forward
    Operators to calculate the loss; second, run backward Operators to 
    calculate gradient of the parameters; third, apply optimization method
    to update the value of the parameters.

    In the forward computation process, all variables that are needed by 
    backward computation process will be kept in memory, which occupy a great
    amount of memory when the network becomes very deep.

    Recompute split the network to k segments. In each segment, It will 
    recompute the forward Operators, before running backward operators. It is
    very helpful for saving memory.
 
    The Variables that separate a network to segments are called as checkpoints,
    and users should set it manually. The usage is very simple:

    Args:
        optimizer (Optimizer): The optimizer that is applied to parameters.

    Examples:
        .. code-block:: python

            import paddle.fluid as fluid
            import numpy as np
            def gen_data():
                return {"x": np.random.random(size=(32, 32)).astype('float32'),
                "y": np.random.randint(2, size=(32, 1)).astype('int64')}
            def mlp(input_x, input_y, hid_dim=128, label_dim=2):
                print(input_x)
                fc_1 = fluid.layers.fc(input=input_x, size=hid_dim)
                prediction = fluid.layers.fc(input=[fc_1], size=label_dim, act='softmax')
                cost = fluid.layers.cross_entropy(input=prediction, label=input_y)
                sum_cost = fluid.layers.reduce_mean(cost)
                return sum_cost, fc_1, prediction
            input_x = fluid.layers.data(name="x", shape=[32], dtype='float32')
            input_y = fluid.layers.data(name="y", shape=[1], dtype='int64')
            cost, fc_1, pred = mlp(input_x, input_y)

            sgd = fluid.optimizer.Adam(learning_rate=0.01)
            sgd = fluid.optimizer.RecomputeOptimizer(sgd)
            sgd._set_checkpoints([fc_1, pred])
            sgd.minimize(cost)

            print("Finished optimize")
            place = fluid.CPUPlace()
            exe = fluid.Executor(place)
            exe.run(fluid.default_startup_program())
            step = 10

            for i in range(step):
                cost_val = exe.run(feed=gen_data(),
                       program=fluid.default_main_program(),
                       fetch_list=[cost.name])
                print("step=%d cost=%f" % (i, cost_val[0]))

    """

    def __init__(self, optimizer):
        self._optimizer = optimizer
        self._checkpoints = None

    def _set_checkpoints(self, checkpoints):
        self._checkpoints = checkpoints

    def load(self, stat_dict):
        """
        load function is not supported by Recompute Optimizer for now.
        :return: None

        Args:
            stat_dict: the dict load by load_persistable method

        Examples:
            .. code-block:: python

                import paddle.fluid as fluid
                import paddle.compat as cpt
                
                def mlp(input_x, input_y, hid_dim=128, label_dim=2):
                    fc_1 = fluid.layers.fc(input=input_x, size=hid_dim)
                    prediction = fluid.layers.fc(input=[fc_1], size=label_dim, act='softmax')
                    cost = fluid.layers.cross_entropy(input=prediction, label=input_y)
                    sum_cost = fluid.layers.reduce_mean(cost)
                    return sum_cost, fc_1, prediction
                
                input_x = fluid.layers.data(name="x", shape=[32], dtype='float32')
                input_y = fluid.layers.data(name="y", shape=[1], dtype='int64')
                cost, fc_1, pred = mlp(input_x, input_y)
                print("Finished FF")
                
                sgd = fluid.optimizer.Adam(learning_rate=0.01)
                sgd = fluid.optimizer.RecomputeOptimizer(sgd)
                sgd._set_checkpoints([fc_1, pred])
                try:
                    stat_dict = {}
                    sgd.load(stat_dict)
                except NotImplementedError as e:
                    print(cpt.get_exception_message(e))
        """
        raise NotImplementedError(
            "load function is not supported by Recompute Optimizer for now")

    def apply_gradients(self, params_grads):
        """
        call apply_gradients function of self._optimizer.

        Args:
            params_grads (list): list of (param, grad) pair to do optimization.

        Returns:
            list: A list of operators appended to the current program.

        Examples:
            .. code-block:: python

                import paddle.fluid as fluid
                import paddle.fluid.framework as framework

                def mlp(input_x, input_y, hid_dim=128, label_dim=2):
                    fc_1 = fluid.layers.fc(input=input_x, size=hid_dim)
                    prediction = fluid.layers.fc(input=[fc_1], size=label_dim, act='softmax')
                    cost = fluid.layers.cross_entropy(input=prediction, label=input_y)
                    sum_cost = fluid.layers.reduce_mean(cost)
                    return sum_cost, fc_1, prediction


                input_x = fluid.layers.data(name="x", shape=[32], dtype='float32')
                input_y = fluid.layers.data(name="y", shape=[1], dtype='int64')
                cost, fc_1, pred = mlp(input_x, input_y)
                print("Finished FF")

                sgd = fluid.optimizer.Adam(learning_rate=0.01)
                sgd = fluid.optimizer.RecomputeOptimizer(sgd)
                params_grads = sgd.backward(
                    cost,
                    startup_program=None,
                    parameter_list=None,
                    no_grad_set=None,
                    checkpoints=[fc_1, pred])

                program = cost.block.program
                with framework.program_guard(program, None):
                    optimize_ops = sgd.apply_gradients(params_grads)

                print("Finished apply gradients")
        """

        return self._optimizer.apply_gradients(params_grads=params_grads)

    def backward(self,
                 loss,
                 startup_program=None,
                 parameter_list=None,
                 no_grad_set=None,
                 callbacks=None,
                 checkpoints=None):
        """
        call append_backward with checkpoints.

        Args:
            loss (Variable): loss variable to run optimizations.
            startup_program (Program): startup_program for initializing parameters
                in `parameter_list`.
            parameter_list (list): list of Variables to update.
            no_grad_set (set|None): set of Variables should be ignored.
            callbacks (list|None): list of callables to run when appending backward
                operator for one parameter.
            checkpoints (list): list of Variables as checkpoints

        Examples:
            .. code-block:: python

                import paddle.fluid as fluid
    
                def mlp(input_x, input_y, hid_dim=128, label_dim=2):
                    fc_1 = fluid.layers.fc(input=input_x, size=hid_dim)
                    prediction = fluid.layers.fc(input=[fc_1], size=label_dim, act='softmax')
                    cost = fluid.layers.cross_entropy(input=prediction, label=input_y)
                    sum_cost = fluid.layers.reduce_mean(cost)
                    return sum_cost, fc_1, prediction
    
    
                input_x = fluid.layers.data(name="x", shape=[32], dtype='float32')
                input_y = fluid.layers.data(name="y", shape=[1], dtype='int64')
                cost, fc_1, pred = mlp(input_x, input_y)
                print("Finished FF")
    
                sgd = fluid.optimizer.Adam(learning_rate=0.01)
                sgd = fluid.optimizer.RecomputeOptimizer(sgd)
                params_grads = sgd.backward(
                    cost,
                    startup_program=None,
                    parameter_list=None,
                    no_grad_set=None,
                    checkpoints=[fc_1, pred])
                print("Finished backward")
        """

        if framework.in_dygraph_mode():
            raise NotImplementedError(
                "DyGraph current does not support recompute")

        self._dtype = loss.dtype
        program = loss.block.program
        with program_guard(program, startup_program):
            params_grads = append_backward(
                loss,
                parameter_list,
                no_grad_set,
                checkpoints=self._checkpoints)
        return params_grads

    def apply_optimize(self, loss, startup_program, params_grads):
        """
        call the apply_optimize function of self._optimizer

        Args:
            loss (Variable): loss variable to run optimizations.
            startup_program (Program): startup_program for initializing parameters
                in `parameter_list`.
            params_grads (list): list of (param, grad) pair to do optimization.

        Examples:
            .. code-block:: python
M
mapingshuo 已提交
3624

M
mapingshuo 已提交
3625 3626 3627 3628 3629 3630 3631
                import paddle.fluid as fluid
                
                def mlp(input_x, input_y, hid_dim=128, label_dim=2):
                    fc_1 = fluid.layers.fc(input=input_x, size=hid_dim)
                    prediction = fluid.layers.fc(input=[fc_1], size=label_dim, act='softmax')
                    cost = fluid.layers.cross_entropy(input=prediction, label=input_y)
                    sum_cost = fluid.layers.reduce_mean(cost)
M
mapingshuo 已提交
3632
                    return sum_cost, fc_1, prediction                
M
mapingshuo 已提交
3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651
                
                input_x = fluid.layers.data(name="x", shape=[32], dtype='float32')
                input_y = fluid.layers.data(name="y", shape=[1], dtype='int64')
                cost, fc_1, pred = mlp(input_x, input_y)
                print("Finished FF")
                
                sgd = fluid.optimizer.Adam(learning_rate=0.01)
                sgd = fluid.optimizer.RecomputeOptimizer(sgd)
                params_grads = sgd.backward(
                    cost,
                    startup_program=None,
                    parameter_list=None,
                    no_grad_set=None,
                    checkpoints=[fc_1, pred])
                
                optimize_ops = sgd.apply_optimize(
                    cost, startup_program=None, params_grads=params_grads)
                
                print("Finished apply_optimize")
M
mapingshuo 已提交
3652

M
mapingshuo 已提交
3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688
        """

        return self._optimizer.apply_optimize(
            loss, startup_program=startup_program, params_grads=params_grads)

    def minimize(self,
                 loss,
                 startup_program=None,
                 parameter_list=None,
                 no_grad_set=None,
                 grad_clip=None):

        assert (isinstance(loss, Variable)), "The loss should be an Variable."
        assert (self._checkpoints is not None
                ), "You should call _set_checkpoints first"
        if framework.in_dygraph_mode():
            raise NotImplementedError(
                "DyGraph current does not support recompute")

        params_grads = self.backward(
            loss,
            startup_program=startup_program,
            parameter_list=parameter_list,
            no_grad_set=no_grad_set,
            checkpoints=self._checkpoints)

        if grad_clip:
            # TODO(guru4elephant): should add grad_clip for static graph
            pass

        optimize_ops = self.apply_optimize(
            loss, startup_program=startup_program, params_grads=params_grads)

        return optimize_ops, params_grads


M
mapingshuo 已提交
3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839
class LookaheadOptimizer(object):
    """
    This implements the Lookahead optimizer of the
    paper : https://arxiv.org/abs/1907.08610.

    Lookahead keeps two sets of params: the fast_params and
    the slow_params. inner_optimizer update fast_params every 
    training step. Lookahead updates the slow_params and fast_params 
    every k training steps as follows:

    .. math::
        
        slow\_param_t &= slow\_param_{t-1} + \\alpha * (fast\_param_{t-1} - slow\_param_{t-1})
	
	fast\_param_t &=  slow\_param_t

    Args:
        inner_optimizer (Optimizer): The optimizer that update fast params step by step. 
        alpha (float): The learning rate of Lookahead.
        k (int): The slow params is updated every k steps.

    Examples:
        .. code-block:: python

            import paddle
            import paddle.fluid as fluid
            import numpy as np

	    x = fluid.layers.data(name='x', shape=[2], dtype='float32')
	    label = fluid.layers.data(name="label", shape=[1], dtype="int64")
	    y = fluid.layers.fc(input=[x], size=2, act="softmax")
	    loss = fluid.layers.cross_entropy(input=y, label=label)
	    loss = fluid.layers.mean(x=loss)
	    sgd = fluid.optimizer.SGD(learning_rate=0.01)
	    optimizer = fluid.optimizer.LookaheadOptimizer(sgd,
                                            alpha=0.5,
                                            k=5)
	    optimizer.minimize(loss)
	    main_program = fluid.default_main_program()
	    place = fluid.CPUPlace()
	    exe = fluid.Executor(place)
	    exe.run(fluid.default_startup_program())

	    feeder = fluid.DataFeeder(feed_list=[x, label], place=place)

	    step = 0
            while(step < 10):
                step += 1
		exe.run(fluid.default_main_program(),
            	feed=feeder.feed(batch_data))

    """

    def __init__(self, inner_optimizer, alpha=0.5, k=5):

        assert (inner_optimizer is not None), "inner optimizer can not be None"
        assert (
            0.0 <= alpha <= 1.0
        ), "alpha should be larger or equal to 0.0, and less or equal than 1.0"
        assert (isinstance(k, int) and k > 0), "k should be a positive integer"

        self.inner_optimizer = inner_optimizer
        self.alpha = alpha
        self.k = k
        self.type = "lookahead"

    def minimize(self, loss, startup_program=None):

        # Apply inner optimizer to the main_program
        mini_out = self.inner_optimizer.minimize(
            loss, startup_program=startup_program)

        # Get startup_program and main_program
        if startup_program is None:
            startup_program = default_startup_program()
        main_block = loss.block

        # add some vars to the main_program
        params = [param.name for param in main_block.all_parameters()]
        param_to_slow = {}
        for param in params:
            fast_var = main_block.var(param)
            assert (fast_var is not None)
            slow_var = main_block.create_var(
                name=param + "@SLOW",
                shape=fast_var.shape,
                dtype=fast_var.dtype,
                persistable=True)
            param_to_slow[param] = slow_var

        # add some vars to the startup_program
        startup_block = startup_program.global_block()
        for param in params:
            fast_var = startup_block.var(param)
            assert (fast_var is not None)
            slow_var = startup_block.create_var(
                name=param + "@SLOW",
                shape=fast_var.shape,
                dtype=fast_var.dtype,
                persistable=True)

            startup_block.append_op(
                type="assign",
                inputs={"X": fast_var},
                outputs={"Out": slow_var})

        # Add Var k to main prog and startup prog
        k = layers.create_global_var(
            name="lookahead_k",
            shape=[1],
            value=int(self.k),
            dtype='int32',
            persistable=True)

        # Add Var alpha to main prog and startup prog
        alpha = layers.create_global_var(
            name="lookahead_alpha",
            shape=[1],
            value=float(self.alpha),
            dtype='float32',
            persistable=True)

        # Add Var step
        step = layers.create_global_var(
            name="lookahead_step",
            shape=[1],
            value=int(0),
            dtype='int32',
            persistable=True)
        layers.increment(x=step, value=1.0, in_place=True)

        # lookahead
        zero_var = layers.fill_constant(shape=[1], dtype='float32', value=0.0)

        one_var = layers.fill_constant(shape=[1], dtype='float32', value=1.0)

        mod = layers.elementwise_mod(step, k)
        with layers.control_flow.Switch() as switch:
            with switch.case(mod == zero_var):
                for param_name in params:
                    fast_var = main_block.var(param_name)
                    slow_var = param_to_slow[param_name]
                    tmp_var = layers.elementwise_add(
                        layers.elementwise_mul(fast_var, alpha),
                        layers.elementwise_mul(
                            slow_var, layers.elementwise_sub(one_var, alpha)))
                    layers.assign(input=tmp_var, output=slow_var)
                    layers.assign(input=tmp_var, output=fast_var)
            with switch.default():
                pass
        return mini_out