optimizer.py 163.4 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
26
from .backward import append_backward, _some_in_set_, _append_grad_suffix_, _get_no_grad_set_name
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
from .dygraph import base as imperative_base
34
from .dygraph import no_grad
35 36 37 38
from .dygraph.learning_rate_scheduler import LearningRateDecay
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
63 64 65 66 67 68
    def __init__(self,
                 learning_rate,
                 parameter_list=None,
                 regularization=None,
                 name=None):
        self._parameter_list = None
L
lujun 已提交
69
        if framework.in_dygraph_mode():
M
minqiyang 已提交
70 71 72 73 74
            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))
75 76 77 78
            if name is not None:
                self._name = unique_name.generate(name)
            else:
                self._name = unique_name.generate(self.__class__.__name__)
79 80 81 82 83 84
            if parameter_list is not None:
                self._parameter_list = parameter_list
            else:
                raise AttributeError(
                    "parameter_list argument given to the Optimizer should not be None in dygraph mode."
                )
M
minqiyang 已提交
85 86 87 88 89 90
        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))
91
            self._name = name
M
minqiyang 已提交
92

D
dzhwinter 已提交
93
        self.regularization = regularization
94
        self._learning_rate = learning_rate
D
dzhwinter 已提交
95 96
        # the learning rate type should be inferenced from loss
        self._dtype = None
97
        # each program should have a independent learning rate
98
        # program -> Variable(learning_rate)
Q
qiaolongfei 已提交
99
        self._learning_rate_map = dict()
100
        if isinstance(self._learning_rate, framework.Variable):
101 102
            self._learning_rate_map[framework.default_main_program(
            )] = self._learning_rate
103 104 105 106 107
        # 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
108
        self.helper = None
109
        self._opti_name_list = []
H
hong 已提交
110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125
        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
126 127 128 129 130 131

                with fluid.dygraph.guard():
                    emb = fluid.dygraph.Embedding([10, 10])

                    adam = fluid.optimizer.Adam(0.001, parameter_list=emb.parameters())
                    state_dict = adam.state_dict()
H
hong 已提交
132 133 134 135 136 137 138 139

        '''
        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):
140
            var_tmp = None
L
Leo Chen 已提交
141
            if framework.in_dygraph_mode():
142 143
                var_temp = framework._varbase_creator(
                    None, name='global_step', dtype='int32')
L
Leo Chen 已提交
144 145
            else:
                var_temp = Variable(None, name='global_step', dtype='int32')
146

H
hong 已提交
147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164
            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
165

H
hong 已提交
166
                with fluid.dygraph.guard():
167
                    emb = fluid.dygraph.Embedding([10, 10])
168

H
hong 已提交
169
                    state_dict = emb.state_dict()
170
                    fluid.save_dygraph(state_dict, "paddle_dy")
171

172 173
                    adam = fluid.optimizer.Adam(learning_rate=fluid.layers.noam_decay( 100, 10000), 
                                                parameter_list=emb.parameters())
H
hong 已提交
174
                    state_dict = adam.state_dict()
175
                    fluid.save_dygraph(state_dict, "paddle_dy")
176

H
hong 已提交
177
                    para_state_dict, opti_state_dict = fluid.load_dygraph( "paddle_dy")
178

179
                    adam.set_dict(opti_state_dict)
H
hong 已提交
180 181 182 183 184 185 186 187 188

        '''

        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):
189
                step_np = global_step
H
hong 已提交
190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213
                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 )
214
                var = var_tmp.value()
H
hong 已提交
215 216 217 218 219 220 221 222
                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):
223
                    load_para_np = load_para.numpy()
H
hong 已提交
224 225 226 227 228 229 230 231 232 233 234 235 236 237 238
                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())
239

240 241
    def get_opti_var_name_list(self):
        return self._opti_name_list
Q
Qiao Longfei 已提交
242

Q
Qiao Longfei 已提交
243
    def _create_global_learning_rate(self):
244 245 246
        if imperative_base.enabled():
            # create learning rate Variable
            if isinstance(self._learning_rate, float):
M
minqiyang 已提交
247 248 249 250 251 252 253 254 255 256 257 258
                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)
259
            # get learning rate Variable from LearningRateDecay
M
minqiyang 已提交
260
            elif isinstance(self._learning_rate, LearningRateDecay):
261 262 263
                self._learning_rate_map[framework.default_main_program(
                )] = self._learning_rate()
            else:
Q
qiaolongfei 已提交
264
                raise TypeError(
265 266
                    "optimizer's learning rate must be float or LearningRateDecay"
                )
267
        else:
268 269 270 271
            lr = self._global_learning_rate()

            if isinstance(lr, framework.Variable):
                return
M
minqiyang 已提交
272 273 274 275 276 277
            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
278

279 280 281 282 283 284 285 286
            # 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)
287

288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349
    @framework.dygraph_only
    def current_step_lr(self):
        """
        .. note::
          **This API is ONLY avaliable in Dygraph mode**
        
        Get current step learning rate. The return value is all the same When LearningRateDecay is not used,
        otherwise return the step learning rate.

        Returns:
            float: The learning rate of the current step.

        Examples:
            .. code-block:: python

                import paddle.fluid as fluid
                import numpy as np

                # example1: LearningRateDecay is not used, return value is all the same
                with fluid.dygraph.guard():
                    emb = fluid.dygraph.Embedding([10, 10])
                    adam = fluid.optimizer.Adam(0.001, parameter_list = emb.parameters())
                    lr = adam.current_step_lr()
                    print(lr) # 0.001

                # example2: PiecewiseDecay is used, return the step learning rate
                with fluid.dygraph.guard():
                    inp = np.random.uniform(-0.1, 0.1, [10, 10]).astype("float32")
                    linear = fluid.dygraph.nn.Linear(10, 10)
                    inp = fluid.dygraph.to_variable(inp)
                    out = linear(inp)
                    loss = fluid.layers.reduce_mean(out)
                    
                    bd = [2, 4, 6, 8]
                    value = [0.2, 0.4, 0.6, 0.8, 1.0]
                    adam = fluid.optimizer.Adam(fluid.dygraph.PiecewiseDecay(bd, value, 0),
                                           parameter_list=linear.parameters())

                    # first step: learning rate is 0.2
                    np.allclose(adam.current_step_lr(), 0.2, rtol=1e-06, atol=0.0) # True

                    # learning rate for different steps
                    ret = [0.2, 0.2, 0.4, 0.4, 0.6, 0.6, 0.8, 0.8, 1.0, 1.0, 1.0, 1.0]
                    for i in range(12):
                        adam.minimize(loss)
                        lr = adam.current_step_lr()
                        np.allclose(lr, ret[i], rtol=1e-06, atol=0.0) # True

        """
        current_lr = self._global_learning_rate()
        if current_lr:
            return self._global_learning_rate().numpy()[0]

        if isinstance(self._learning_rate, float):
            return self._learning_rate
        else:
            step_lr = self._learning_rate.step()
            if isinstance(step_lr, (float, int)):
                return step_lr
            else:
                return step_lr.numpy()[0]

Y
yuyang18 已提交
350
    def _global_learning_rate(self, program=None):
Q
Qiao Longfei 已提交
351 352 353 354
        """
        get global decayed learning rate
        :return:
        """
355 356
        if program is None:
            program = framework.default_main_program()
Q
qiaolongfei 已提交
357
        return self._learning_rate_map.get(program, None)
Q
Qiao Longfei 已提交
358

Q
Qiao Longfei 已提交
359 360 361 362 363
    def _append_optimize_op(self, block, param_and_grad):
        """ append optimize operator to block and return all the added optimize_op
        """
        raise NotImplementedError()

364 365 366 367
    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
368 369
        if type(param_lr) == Variable:
            return param_lr
Q
qiaolongfei 已提交
370
        else:
W
Wu Yi 已提交
371
            if param_lr == 1.0:
Y
yuyang18 已提交
372
                return self._global_learning_rate()
W
Wu Yi 已提交
373
            else:
X
Xin Pan 已提交
374 375 376
                with default_main_program()._lr_schedule_guard(
                        is_with_opt=True), framework.name_scope(
                            'scale_with_param_lr'):
377
                    return self._global_learning_rate() * param_lr
378 379 380 381 382 383 384

    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
385
        """
386 387
        pass

388
    def _finish_update(self, block, parameters_and_grads):
389 390 391 392 393 394 395 396
        """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
397
            None
398 399 400
        """
        pass

401 402 403 404 405
    def _add_accumulator(self,
                         name,
                         param,
                         dtype=None,
                         fill_value=0.0,
406 407
                         shape=None,
                         type=None):
408 409 410 411 412 413 414 415 416
        """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
417 418
        if self._name is not None:
            name = self._name + "_" + name
419 420
        if (name in self._accumulators and
                param.name in self._accumulators[name]):
L
lujun 已提交
421
            if framework.in_dygraph_mode():
X
polish  
Xin Pan 已提交
422
                return self._accumulators[name][param.name]
423
            raise Exception("Accumulator {} already exists for parameter {}".
424
                            format(name, param.name))
425 426
        if shape == None:
            shape = param.shape
Q
Qiao Longfei 已提交
427
        assert isinstance(self.helper, LayerHelper)
428 429 430 431 432

        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
433
        var = self.helper.create_global_variable(
434
            name=var_name,
Q
Qiao Longfei 已提交
435
            persistable=True,
F
fengjiayi 已提交
436
            dtype=dtype or param.dtype,
437
            type=param.type if type is None else type,
H
hong 已提交
438 439
            shape=shape,
            belong_to_optimizer=True)
Q
Qiao Longfei 已提交
440
        self.helper.set_variable_initializer(
441
            var, initializer=Constant(value=float(fill_value)))
H
hong 已提交
442 443 444 445 446 447 448

        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
449
        self._accumulators[name][param.name] = var
450
        return var
451 452 453 454 455 456 457 458 459 460 461

    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
462 463
        if self._name is not None:
            name = self._name + "_" + name
464 465 466 467 468 469
        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]

470
    def _create_optimization_pass(self, parameters_and_grads):
Q
Qiao Longfei 已提交
471 472 473
        """Add optimization operators to update gradients to variables.

        Args:
Q
qiaolongfei 已提交
474
          parameters_and_grads(list(tuple(Variable, Variable))):
475
            a list of (variable, gradient) pair to update.
Q
Qiao Longfei 已提交
476 477

        Returns:
478
          return_op_list: a list of operators that will complete one step of
479 480 481
            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
482
        """
483 484 485 486 487
        # 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
488
        # for parameters and extend _finish_update method to add custom ops.
489

490
        # Allways called under program_guard use global block as loss block
491 492 493
        # But if current block is in control flow, append optimize op in the
        # grad block of current block

494
        global_block = framework.default_main_program().global_block()
495 496 497 498 499 500 501 502 503
        target_block = global_block
        current_block = framework.default_main_program().current_block()
        if current_block.idx != global_block.idx:
            assert current_block.backward_block_idx != -1, \
                "current block is not global_block, but it doesn't have backward block."
            target_block = framework.default_main_program().blocks[
                current_block.backward_block_idx]

        start = len(target_block.ops)
504
        self.helper = LayerHelper(self.__class__.__name__)
C
chengduo 已提交
505
        self._create_accumulators(
506
            target_block,
C
chengduo 已提交
507
            [p[0] for p in parameters_and_grads if p[0].trainable])
508 509
        self._create_global_learning_rate()

M
minqiyang 已提交
510
        if framework.in_dygraph_mode():
511 512 513
            for param_and_grad in parameters_and_grads:
                if param_and_grad[1] is None:
                    continue
514 515
                if param_and_grad[0].trainable is True:
                    self._append_optimize_op(target_block, param_and_grad)
516 517 518 519 520 521 522
        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:
523
                        self._append_optimize_op(target_block, param_and_grad)
524 525 526

        # Get custom finish ops for subclasses
        # FIXME: Need to fix this once we figure out how to handle dependencies
527
        self._finish_update(target_block, parameters_and_grads)
528

529 530
        end = len(target_block.ops)
        return target_block._slice_ops(start, end)
531 532

    def _process_distribute_lookuptable(self, param_grads):
Q
Qiao Longfei 已提交
533 534 535 536 537 538 539 540 541
        """
        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
        """
542 543
        program = framework.default_main_program()
        global_block = framework.default_main_program().global_block()
Q
Qiao Longfei 已提交
544 545 546 547 548 549 550 551 552 553 554 555 556 557 558
        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:
559 560 561 562 563 564 565 566 567 568 569 570 571
            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
572 573
        return new_param_grads, (table_param, table_grad), sgd_op

574 575 576
    def _append_dgc_ops(self, param_and_grad):
        pass

577 578 579 580 581 582 583
    def backward(self,
                 loss,
                 startup_program=None,
                 parameter_list=None,
                 no_grad_set=None,
                 callbacks=None):
        """
584
        The first part of ``minimize``, do auto-diff to append backward operations for
585 586 587
        the current program.

        Args:
588 589 590 591
            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.
592
            parameter_list (list, optional): List of ``Variable`` or ``Variable.name`` to update
593 594
                to minimize ``loss``. The default value is None, at this time all parameters
                will be updated.
595
            no_grad_set (set, optional): Set of ``Variable``  or ``Variable.name`` that don't need
596 597 598
                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
599

600
        Return:
601 602
            list: list of (param, grad) variable pairs, param is ``Parameter``,
                grad is the gradient value corresponding to the parameter.
M
minqiyang 已提交
603

604
        Examples:
605
            See examples in ``apply_gradients``.
606
        """
607
        act_no_grad_set = None
L
Leo Chen 已提交
608
        if framework.in_dygraph_mode():
609
            pass
L
Leo Chen 已提交
610 611
        else:
            act_no_grad_set = self._get_no_grad_set(loss, no_grad_set)
G
gongweibao 已提交
612

C
chengduo 已提交
613
        self._dtype = loss.dtype
L
lujun 已提交
614
        if framework.in_dygraph_mode():
C
chengduo 已提交
615
            params_grads = []
616
            for param in self._parameter_list:
C
chengduo 已提交
617 618
                if not param.trainable:
                    continue
619
                if param._grad_ivar() is not None:
C
chengduo 已提交
620
                    # create gradient variable
621
                    grad_var = param._grad_ivar()
C
chengduo 已提交
622
                    params_grads.append((param, grad_var))
623
        else:
C
chengduo 已提交
624 625 626 627 628
            if callbacks is None:
                callbacks = [error_clip_callback]
            else:
                assert (isinstance(callbacks, list))
            program = loss.block.program
C
chengduo 已提交
629 630 631 632
            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
633 634
            with program_guard(program, startup_program):
                params_grads = append_backward(loss, parameter_list,
635
                                               act_no_grad_set, callbacks)
C
chengduo 已提交
636 637 638 639
                # 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
640 641 642 643 644 645 646 647

    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
648

649 650
        Returns:
            list: A list of operators appended to the current program.
M
minqiyang 已提交
651

652 653 654
        Examples:
            .. code-block:: python

655
                import paddle.fluid as fluid
656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680
                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
681 682 683 684 685 686 687 688 689 690 691 692 693 694
    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
695
        if framework.in_dygraph_mode():
C
chengduo 已提交
696 697
            with program_guard(framework.default_main_program(),
                               framework.default_startup_program()):
698 699
                params_grads = append_regularization_ops(params_grads,
                                                         self.regularization)
C
chengduo 已提交
700 701 702 703 704 705 706
                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
707
    def _get_no_grad_set(self, loss, no_grad_set=None):
708
        no_grad_set = _get_no_grad_set_name(no_grad_set)
G
gongweibao 已提交
709 710 711 712 713 714 715 716
        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

717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747
    @framework.dygraph_only
    def clear_gradients(self):
        """
        Clear the gradients of all optimized parameters for model.
        
        Returns:
            None
        
        Examples:
            .. code-block:: python

                import paddle.fluid as fluid
                import numpy as np

                with fluid.dygraph.guard():
                    value = np.arange(26).reshape(2, 13).astype("float32")
                    a = fluid.dygraph.to_variable(value)
                    linear = fluid.Linear(13, 5, dtype="float32")
                    # This can be any optimizer supported by dygraph.
                    adam = fluid.optimizer.Adam(learning_rate = 0.01, 
                                                parameter_list = linear.parameters())
                    out = linear(a)
                    out.backward()
                    adam.minimize(out)
                    adam.clear_gradients()

        """
        for p in self._parameter_list:
            if p.trainable:
                p.clear_gradient()

748
    @imperative_base.no_grad
Q
Qiao Longfei 已提交
749 750
    def minimize(self,
                 loss,
751
                 startup_program=None,
Q
Qiao Longfei 已提交
752
                 parameter_list=None,
753 754
                 no_grad_set=None,
                 grad_clip=None):
755
        """
756
        Add operations to minimize ``loss`` by updating ``parameter_list``.
M
minqiyang 已提交
757

758
        Args:
759 760 761 762
            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.
763
            parameter_list (list, optional): List of ``Variable`` or ``Variable.name`` to update
764 765
                to minimize ``loss``. The default value is None, at this time all parameters
                will be updated.
766
            no_grad_set (set, optional): Set of ``Variable``  or ``Variable.name`` that don't need
767 768 769 770 771
                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
772

773
        Returns:
774 775 776 777 778 779
            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
780
        """
C
chengduo 已提交
781
        assert isinstance(loss, Variable), "The loss should be an Variable."
C
chengduo 已提交
782 783 784 785 786
        params_grads = self.backward(
            loss,
            startup_program=startup_program,
            parameter_list=parameter_list,
            no_grad_set=no_grad_set)
787 788 789 790 791

        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
792 793
        optimize_ops = self.apply_optimize(
            loss, startup_program=startup_program, params_grads=params_grads)
M
minqiyang 已提交
794

Q
Qiao Longfei 已提交
795
        return optimize_ops, params_grads
Q
Qiao Longfei 已提交
796 797 798


class SGDOptimizer(Optimizer):
Q
qiaolongfei 已提交
799 800 801 802 803 804 805
    """
    Optimizer of the stochastic gradient descent algorithm.

    .. math::

        param\_out = param - learning\_rate * grad

806 807 808
    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.
809 810 811
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
812 813 814 815
        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
816 817 818 819

    Examples:
        .. code-block:: python

820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844
            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
845 846
    """

847 848 849 850 851
    def __init__(self,
                 learning_rate,
                 parameter_list=None,
                 regularization=None,
                 name=None):
Q
Qiao Longfei 已提交
852
        assert learning_rate is not None
Q
Qiao Longfei 已提交
853
        super(SGDOptimizer, self).__init__(
X
Xin Pan 已提交
854
            learning_rate=learning_rate,
855
            parameter_list=parameter_list,
X
Xin Pan 已提交
856 857
            regularization=regularization,
            name=name)
Q
Qiao Longfei 已提交
858 859
        self.type = "sgd"

860
    @no_grad
861
    def _append_optimize_op(self, block, param_and_grad):
862 863 864 865 866 867 868 869 870 871
        if framework.in_dygraph_mode():
            inputs = {
                "Param": [param_and_grad[0]],
                "Grad": [param_and_grad[1]],
                "LearningRate": [self._create_param_lr(param_and_grad)]
            }
            attrs = {}
            outputs = {'ParamOut': [param_and_grad[0]]}
            outs = core.ops.sgd(inputs, attrs, outputs)
            return outs['ParamOut'][0]
872

873
        assert isinstance(block, framework.Block)
Q
Qiao Longfei 已提交
874 875 876 877 878 879
        # create the optimize op
        sgd_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
880
                "LearningRate": self._create_param_lr(param_and_grad)
Q
Qiao Longfei 已提交
881
            },
M
minqiyang 已提交
882 883
            outputs={"ParamOut": param_and_grad[0]},
            stop_gradient=True)
Q
Qiao Longfei 已提交
884 885

        return sgd_op
886 887 888


class MomentumOptimizer(Optimizer):
Q
qiaolongfei 已提交
889 890 891 892 893 894 895 896 897 898 899 900 901 902
    """

    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):

903
        &\quad   param = param - (gradient + mu * velocity) * learning\_rate
Q
qiaolongfei 已提交
904 905 906

        & else:

Q
qiaolongfei 已提交
907
        &\quad   param = param - learning\_rate * velocity
Q
qiaolongfei 已提交
908

909 910 911 912
    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
913 914 915
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
916 917 918 919 920
        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
921 922 923 924

    Examples:
        .. code-block:: python

925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949
            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)

950 951 952
    """
    _velocity_acc_str = "velocity"

X
Xin Pan 已提交
953 954 955
    def __init__(self,
                 learning_rate,
                 momentum,
956
                 parameter_list=None,
X
Xin Pan 已提交
957 958 959
                 use_nesterov=False,
                 regularization=None,
                 name=None):
960 961
        assert learning_rate is not None
        assert momentum is not None
Q
Qiao Longfei 已提交
962
        super(MomentumOptimizer, self).__init__(
X
Xin Pan 已提交
963
            learning_rate=learning_rate,
964
            parameter_list=parameter_list,
X
Xin Pan 已提交
965 966
            regularization=regularization,
            name=name)
967 968
        self.type = "momentum"
        self._momentum = momentum
969
        self._use_nesterov = bool(use_nesterov)
970 971 972 973 974

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

        for p in parameters:
Q
Qiao Longfei 已提交
975
            self._add_accumulator(self._velocity_acc_str, p)
976 977 978 979 980 981

    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])
982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999
        attrs = {"mu": self._momentum, "use_nesterov": self._use_nesterov}

        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]
        }

        if framework.in_dygraph_mode():
            core.ops.momentum(inputs, attrs, outputs)
            return None

1000 1001 1002
        # create the momentum optimize op
        momentum_op = block.append_op(
            type=self.type,
1003 1004 1005
            inputs=inputs,
            outputs=outputs,
            attrs=attrs,
M
minqiyang 已提交
1006
            stop_gradient=True)
1007 1008

        return momentum_op
1009 1010


1011
class DGCMomentumOptimizer(Optimizer):
1012
    """
1013
    DGC (Deep Gradient Compression) Momentum Optimizer. Original paper is https://arxiv.org/abs/1712.01887
1014

G
gongweibao 已提交
1015
    DGC reduces the communication bandwidth by sending only the important gradients (sparse update):\
1016 1017
        only gradients larger than a threshold are transmitted.

G
gongweibao 已提交
1018
    To avoid losing information, DGC accumulates the rest of the gradients locally.
1019 1020 1021

    Eventually, these gradients become large enough to be transmitted.

1022
    Thus, DGC sends the large gradients immediately but eventually sends all of the gradients over time.
1023

G
gongweibao 已提交
1024
    To ensure no loss of accuracy, DGC employs momentum correction and local gradient clipping on top of the gradient sparsification to maintain model performance.
1025 1026 1027 1028

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

    This optimizer will do two things:
1029

1030 1031
        1. Compress the gradient by get TopK import value from tensor \
            and use it for allreduce to reduce network bandwidth.
1032

1033
        2. Call momentum to optimize the cost.
1034 1035

    Args:
1036 1037
        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.
1038
        momentum (float): Momentum factor.
G
gongweibao 已提交
1039
        rampup_begin_step (int): The beginning step from which gradient compression is implemented.
1040 1041 1042 1043 1044 1045 1046
        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.
1047 1048 1049
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
1050 1051 1052 1053 1054 1055 1056
        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.
1057 1058 1059 1060

    Examples:
        .. code-block:: python

1061
            import paddle.fluid as fluid
1062
            optimizer = fluid.optimizer.DGCMomentumOptimizer(
G
gongweibao 已提交
1063 1064 1065 1066 1067
                        learning_rate=0.0001,
                        momentum=0.9,
                        rampup_step=1000,
                        rampup_begin_step=1252,
                        sparsity=[0.999, 0.999])
1068 1069

    """
1070 1071
    _u_velocity_acc_str = "_dgc_u_"
    _v_velocity_acc_str = "_dgc_v_"
1072 1073 1074 1075 1076 1077 1078

    def __init__(self,
                 learning_rate,
                 momentum,
                 rampup_begin_step,
                 rampup_step=1,
                 sparsity=[0.999],
1079
                 parameter_list=None,
1080 1081 1082 1083 1084
                 use_nesterov=False,
                 local_grad_clip_norm=None,
                 num_trainers=None,
                 regularization=None,
                 name=None):
Z
zhongpu 已提交
1085 1086
        if framework.in_dygraph_mode():
            raise Exception("In dygraph, don't support DGCMomentumOptimizer.")
1087 1088 1089 1090
        assert learning_rate is not None
        assert momentum is not None
        super(DGCMomentumOptimizer, self).__init__(
            learning_rate=learning_rate,
1091
            parameter_list=parameter_list,
1092 1093 1094 1095 1096
            regularization=regularization,
            name=name)
        self.type = "dgc_momentum"
        self._momentum = momentum
        self._use_nesterov = bool(use_nesterov)
1097

1098
        assert rampup_begin_step >= 0, "rampup_begin_step must >= 0"
1099
        self._rampup_begin_step = rampup_begin_step
1100 1101
        self._rampup_step = rampup_step
        self._sparsity = sparsity
1102

1103
        self._rampup_begin_step_var = None
1104
        self._global_step_var = None
1105

1106 1107 1108 1109 1110 1111 1112 1113 1114
        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
1115
            self._clip_norm = local_grad_clip_norm * (num_trainers**-0.5)
1116

1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132
        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'

1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143
    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)
1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158
        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}
1159 1160

        if not self._is_use_dgc(param_and_grad[0], param_and_grad[1]):
1161 1162 1163
            type = "momentum"
        else:
            type = "dgc_momentum"
1164 1165 1166 1167 1168
            inputs.update({
                "current_step": self._global_step_var,
                "nranks": self._nranks_var
            })
            outputs.update({'Grad_out': param_and_grad[1]})
1169
            attrs.update({"rampup_begin_step": float(self._rampup_begin_step)})
1170 1171 1172

        # create the dgc momentum optimize op
        dgc_momentum_op = block.append_op(
1173 1174 1175 1176
            type=type,
            inputs=inputs,
            outputs=outputs,
            attrs=attrs,
1177 1178 1179
            stop_gradient=True)
        return dgc_momentum_op

1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198
    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

1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211
    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

1212 1213 1214 1215 1216 1217
    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
1218
            counter_name=core.dgc.kDGCCounterName(), begin=0)
1219

1220 1221 1222
        self._nranks_var = self._add_nranks_var(
            name=core.dgc.kDGCNRanksName(), value=-1)

1223 1224 1225 1226 1227
        # 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
1228
            name=core.dgc.kDGCRampUpBeginStepName(),
1229 1230 1231
            value=self._rampup_begin_step * 1.0,
            force_cpu=True)

1232 1233
        self.helper = LayerHelper(self.__class__.__name__)

1234
        for param_var, grad_var in param_and_grads:
1235 1236 1237
            # reuse velocity in dgc_op and dgc_momentum_op
            u_var = self._add_accumulator(self._u_velocity_acc_str, param_var)

1238
            if not self._is_use_dgc(param_var, grad_var):
1239 1240
                continue

1241
            v_var = self._add_accumulator(self._v_velocity_acc_str, param_var)
1242 1243 1244 1245 1246

            k_var = tensor.create_global_var(
                shape=[1],
                dtype=param_var.dtype,
                persistable=True,
G
gongweibao 已提交
1247
                name=param_var.name + core.dgc.kDGCKName(),
1248 1249 1250 1251 1252 1253 1254
                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
1255
                name=param_var.name + core.dgc.kDGCEncodedName(),
1256 1257 1258
                value=0.0,
                force_cpu=False)

1259 1260 1261 1262 1263 1264 1265 1266
            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)

1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288
            # 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,
1289
                         encoded_var, gather_var)
1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304

    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:
1305 1306
            name = unique_name.generate_with_ignorable_key(".".join(
                [helper.name, 'tmp']))
1307 1308 1309 1310 1311

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

        helper.append_op(
G
gongweibao 已提交
1312
            type="dgc_clip_by_norm",
1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324
            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
1325
                x=grad_var, max_norm=clip_norm, name=grad_var.name)
1326 1327

    def _dgc_op(self, param_var, clip_var, grad_var, u_var, v_var, k_var,
1328
                encoded_var, gather_var):
1329 1330
        block = framework.default_main_program().global_block()
        op_maker = core.op_proto_and_checker_maker
1331

1332 1333 1334 1335 1336 1337
        dgc_op = block.append_op(
            type="dgc",
            inputs={
                "U": u_var,
                "V": v_var,
                "Grad": clip_var,
1338
                "Param": param_var,
1339 1340
                "current_step": self._global_step_var,
                "nranks": self._nranks_var,
1341 1342 1343 1344 1345 1346
            },
            outputs={
                "U_out": u_var,
                "V_out": v_var,
                "EncodeGrad": encoded_var,
                "k": k_var,
1347 1348
                "Grad_out": grad_var,
                "GatherBuff": gather_var,
1349 1350 1351 1352 1353 1354
            },
            attrs={
                "m": self._momentum,
                "sparsity": self._sparsity,
                "use_nesterov": self._use_nesterov,
                "rampup_begin_step": float(self._rampup_begin_step),
1355
                "rampup_step": float(self._rampup_step),
1356 1357
                "regular_coeff": float(self.regular_coeff),
                "regular_type": int(self.regular_type),
1358 1359 1360 1361 1362 1363 1364 1365
            },
            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])

1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396
    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

1397

1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412
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

1413 1414 1415 1416 1417 1418
    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.
1419 1420 1421
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
1422 1423 1424 1425
        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.
1426 1427 1428 1429

    Examples:
        .. code-block:: python

1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445
            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])
1446 1447 1448 1449 1450 1451 1452 1453
    """
    _velocity_acc_str = "velocity"

    def __init__(self,
                 learning_rate,
                 momentum,
                 lars_coeff=0.001,
                 lars_weight_decay=0.0005,
1454
                 parameter_list=None,
1455 1456 1457 1458 1459 1460
                 regularization=None,
                 name=None):
        assert learning_rate is not None
        assert momentum is not None
        super(LarsMomentumOptimizer, self).__init__(
            learning_rate=learning_rate,
1461
            parameter_list=parameter_list,
1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496
            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
1497 1498
            },
            stop_gradient=True)
1499 1500 1501 1502

        return momentum_op


1503
class AdagradOptimizer(Optimizer):
Q
qiaolongfei 已提交
1504
    """
1505 1506
    The Adaptive Gradient optimizer (Adagrad for short) can adaptively assign
    different learning rates to individual parameters.
Q
qiaolongfei 已提交
1507

1508
    The parameter ``param_out`` update rule with gradient ``grad``:
Q
qiaolongfei 已提交
1509 1510 1511 1512 1513 1514 1515

    .. math::

        moment\_out &= moment + grad * grad

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

1516 1517 1518 1519 1520 1521
    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
1522 1523 1524
    for numerical stability to avoid the division by zero error.

    Args:
1525 1526 1527 1528
        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.
1529 1530 1531
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
1532 1533 1534 1535 1536 1537 1538
        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
1539 1540 1541 1542

    Examples:
        .. code-block:: python

1543
            import numpy as np
1544
            import paddle.fluid as fluid
1545 1546

            np_inp = np.array([[1.0, 2.0], [3.0, 4.0]], dtype=np.float32)
1547
            inp = fluid.data(name="inp", shape=[2, 2])
1548 1549
            out = fluid.layers.fc(inp, size=3)
            out = fluid.layers.reduce_sum(out)
1550
            optimizer = fluid.optimizer.AdagradOptimizer(learning_rate=0.2)
1551 1552 1553 1554 1555 1556 1557
            optimizer.minimize(out)

            exe = fluid.Executor(fluid.CPUPlace())
            exe.run(fluid.default_startup_program())
            exe.run(
                feed={"inp": np_inp},
                fetch_list=[out.name])
1558 1559 1560
    """
    _moment_acc_str = "moment"

X
Xin Pan 已提交
1561 1562 1563
    def __init__(self,
                 learning_rate,
                 epsilon=1.0e-6,
1564
                 parameter_list=None,
X
Xin Pan 已提交
1565
                 regularization=None,
1566
                 name=None,
X
xuezhong 已提交
1567
                 initial_accumulator_value=0.0):
1568 1569
        assert learning_rate is not None
        assert epsilon is not None
Q
Qiao Longfei 已提交
1570
        super(AdagradOptimizer, self).__init__(
X
Xin Pan 已提交
1571
            learning_rate=learning_rate,
1572
            parameter_list=parameter_list,
X
Xin Pan 已提交
1573 1574
            regularization=regularization,
            name=name)
1575 1576
        self.type = "adagrad"
        self._epsilon = epsilon
1577
        self.initial_accumulator_value = initial_accumulator_value
1578 1579 1580 1581 1582

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

        for p in parameters:
Z
zhongpu 已提交
1583 1584 1585 1586
            self._add_accumulator(
                self._moment_acc_str,
                p,
                fill_value=self.initial_accumulator_value)
1587 1588 1589 1590 1591 1592

    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])
1593
        # Create the adagrad optimizer op
1594 1595 1596 1597 1598 1599
        adagrad_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "Moment": moment_acc,
1600
                "LearningRate": self._create_param_lr(param_and_grad)
1601 1602 1603
            },
            outputs={"ParamOut": param_and_grad[0],
                     "MomentOut": moment_acc},
M
minqiyang 已提交
1604 1605
            attrs={"epsilon": self._epsilon},
            stop_gradient=True)
1606 1607

        return adagrad_op
1608 1609 1610


class AdamOptimizer(Optimizer):
Q
qiaolongfei 已提交
1611
    """
1612 1613 1614 1615 1616 1617
    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
1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631

    .. 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}

1632 1633
    Related paper: `Adam: A Method for Stochastic Optimization <https://arxiv.org/abs/1412.6980>`_

Q
qiaolongfei 已提交
1634
    Args:
1635 1636
        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.
1637 1638
        beta1 (float|Variable, optional): The exponential decay rate for the 1st moment estimates.
            It should be a float number or a Variable with shape [1] and data type as float32.
1639
            The default value is 0.9.
1640 1641
        beta2 (float|Variable, optional): The exponential decay rate for the 2nd moment estimates.
            It should be a float number or a Variable with shape [1] and data type as float32.
1642 1643 1644
            The default value is 0.999.
        epsilon (float, optional): A small float value for numerical stability.
            The default value is 1e-08.
1645 1646 1647
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659
        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
1660 1661 1662 1663

    Examples:
        .. code-block:: python

1664 1665 1666 1667 1668 1669
            import paddle
            import paddle.fluid as fluid

            place = fluid.CPUPlace()
            main = fluid.Program()
            with fluid.program_guard(main):
1670 1671
                x = fluid.data(name='x', shape=[None, 13], dtype='float32')
                y = fluid.data(name='y', shape=[None, 1], dtype='float32')
1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686
                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
1687

1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704
        .. code-block:: python

            # Adam with beta1/beta2 as Variable
            import paddle
            import paddle.fluid as fluid
            import paddle.fluid.layers.learning_rate_scheduler as lr_scheduler

            place = fluid.CPUPlace()
            main = fluid.Program()
            with fluid.program_guard(main):
                x = fluid.data(name='x', shape=[None, 13], dtype='float32')
                y = fluid.data(name='y', shape=[None, 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)

                # define beta decay variable
1705
                def get_decayed_betas(beta1_init, beta2_init, decay_steps, decay_rate):
1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733
                    global_step = lr_scheduler._decay_step_counter()

                    beta1 = fluid.layers.create_global_var(
                        shape=[1],
                        value=float(beta1_init),
                        dtype='float32',
                        # set persistable for save checkpoints and resume
                        persistable=True,
                        name="beta1")
                    beta2 = fluid.layers.create_global_var(
                        shape=[1],
                        value=float(beta2_init),
                        dtype='float32',
                        # set persistable for save checkpoints and resume
                        persistable=True,
                        name="beta2")

                    div_res = global_step / decay_steps
                    decayed_beta1 = beta1_init * (decay_rate**div_res)
                    decayed_beta2 = beta2_init * (decay_rate**div_res)
                    fluid.layers.assign(decayed_beta1, beta1)
                    fluid.layers.assign(decayed_beta2, beta2)

                    return beta1, beta2

                beta1, beta2 = get_decayed_betas(0.9, 0.99, 1e5, 0.9)
                adam_optimizer = fluid.optimizer.AdamOptimizer(
                                                    learning_rate=0.01,
1734
                                                    beta1=beta1,
1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745
                                                    beta2=beta2)
                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)
1746 1747 1748
    """
    _moment1_acc_str = "moment1"
    _moment2_acc_str = "moment2"
Q
qiaolongfei 已提交
1749 1750
    _beta1_pow_acc_str = "beta1_pow_acc"
    _beta2_pow_acc_str = "beta2_pow_acc"
1751 1752 1753 1754 1755

    def __init__(self,
                 learning_rate=0.001,
                 beta1=0.9,
                 beta2=0.999,
1756
                 epsilon=1e-8,
1757
                 parameter_list=None,
X
Xin Pan 已提交
1758
                 regularization=None,
Q
Qiao Longfei 已提交
1759
                 name=None,
Q
Qiao Longfei 已提交
1760
                 lazy_mode=False):
1761 1762 1763 1764
        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
1765
        super(AdamOptimizer, self).__init__(
X
Xin Pan 已提交
1766
            learning_rate=learning_rate,
1767
            parameter_list=parameter_list,
X
Xin Pan 已提交
1768 1769
            regularization=regularization,
            name=name)
1770 1771 1772 1773
        self.type = "adam"
        self._beta1 = beta1
        self._beta2 = beta2
        self._epsilon = epsilon
Q
Qiao Longfei 已提交
1774
        self._lazy_mode = lazy_mode
1775 1776 1777 1778 1779 1780

    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
1781 1782
            self._add_accumulator(self._moment1_acc_str, p)
            self._add_accumulator(self._moment2_acc_str, p)
Q
qiaolongfei 已提交
1783 1784 1785
            self._add_accumulator(
                name=self._beta1_pow_acc_str,
                param=p,
1786 1787
                fill_value=0.9 if isinstance(self._beta1, Variable) \
                        else self._beta1,
1788 1789
                shape=[1],
                type=core.VarDesc.VarType.LOD_TENSOR)
Q
qiaolongfei 已提交
1790 1791 1792
            self._add_accumulator(
                name=self._beta2_pow_acc_str,
                param=p,
1793 1794
                fill_value=0.999 if isinstance(self._beta2, Variable) \
                        else self._beta2,
1795 1796
                shape=[1],
                type=core.VarDesc.VarType.LOD_TENSOR)
1797 1798 1799 1800 1801 1802 1803 1804

    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
1805 1806 1807 1808 1809
        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])

1810
        # create the adam optimize op
1811
        inputs = {
1812 1813 1814 1815 1816 1817 1818
            "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]
1819 1820
        }
        outputs = {
1821 1822 1823 1824 1825
            "ParamOut": [param_and_grad[0]],
            "Moment1Out": [moment1],
            "Moment2Out": [moment2],
            "Beta1PowOut": [beta1_pow_acc],
            "Beta2PowOut": [beta2_pow_acc],
1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841
        }
        attrs = {
            "epsilon": self._epsilon,
            "lazy_mode": self._lazy_mode,
            "min_row_size_to_use_multithread": 1000
        }

        if isinstance(self._beta1, Variable):
            inputs['Beta1Tensor'] = self._beta1
        else:
            attrs['beta1'] = self._beta1
        if isinstance(self._beta2, Variable):
            inputs['Beta2Tensor'] = self._beta2
        else:
            attrs['beta2'] = self._beta2

1842 1843 1844 1845
        if framework.in_dygraph_mode():
            core.ops.adam(inputs, attrs, outputs)
            return None

1846 1847
        adam_op = block.append_op(
            type=self.type,
1848 1849 1850
            inputs=inputs,
            outputs=outputs,
            attrs=attrs,
M
minqiyang 已提交
1851
            stop_gradient=True)
1852 1853 1854

        return adam_op

1855 1856

class AdamaxOptimizer(Optimizer):
Q
qiaolongfei 已提交
1857
    """
1858 1859 1860 1861
    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
1862

1863
    The parameter ``param_out`` update rule with gradient ``grad``:
Q
qiaolongfei 已提交
1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876

    .. 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}

1877
    Related paper: `Adam: A Method for Stochastic Optimization <https://arxiv.org/abs/1412.6980>`_
Q
qiaolongfei 已提交
1878

1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890
    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.
1891 1892 1893
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
1894 1895 1896 1897 1898 1899 1900 1901
        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
1902

1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915
    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):
1916
              data = fluid.data(name='X', shape=[None, 1], dtype='float32')
1917 1918
              hidden = fluid.layers.fc(input=data, size=10)
              loss = fluid.layers.mean(hidden)
1919
              adam = fluid.optimizer.AdamaxOptimizer(learning_rate=0.2)
1920 1921 1922 1923 1924 1925 1926 1927 1928
              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])
1929 1930 1931
    """
    _moment_acc_str = "moment"
    _inf_norm_acc_str = "inf_norm"
Q
qiaolongfei 已提交
1932
    _beta1_pow_acc_str = "beta1_pow_acc"
1933 1934 1935 1936 1937

    def __init__(self,
                 learning_rate=0.001,
                 beta1=0.9,
                 beta2=0.999,
1938
                 epsilon=1e-8,
1939
                 parameter_list=None,
X
Xin Pan 已提交
1940 1941
                 regularization=None,
                 name=None):
1942 1943 1944 1945
        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
1946
        super(AdamaxOptimizer, self).__init__(
X
Xin Pan 已提交
1947
            learning_rate=learning_rate,
1948
            parameter_list=parameter_list,
X
Xin Pan 已提交
1949 1950
            regularization=regularization,
            name=name)
1951 1952 1953 1954 1955 1956 1957 1958
        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
1959 1960
            self._add_accumulator(self._moment_acc_str, p)
            self._add_accumulator(self._inf_norm_acc_str, p)
Q
qiaolongfei 已提交
1961 1962 1963 1964 1965
            self._add_accumulator(
                name=self._beta1_pow_acc_str,
                param=p,
                fill_value=self._beta1,
                shape=[1])
1966 1967 1968 1969 1970 1971 1972

    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
1973 1974
        beta1_pow_acc = self._get_accumulator(self._beta1_pow_acc_str,
                                              param_and_grad[0])
1975 1976 1977 1978 1979 1980
        # create the adamax optimize op
        adamax_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
1981
                "LearningRate": self._create_param_lr(param_and_grad),
1982 1983
                "Moment": moment,
                "InfNorm": inf_norm,
Q
qiaolongfei 已提交
1984
                "Beta1Pow": beta1_pow_acc
1985 1986 1987 1988 1989 1990 1991 1992 1993 1994
            },
            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
1995 1996
            },
            stop_gradient=True)
1997 1998 1999

        return adamax_op

2000
    def _finish_update(self, block, parameters_and_grads):
2001 2002 2003
        """Update Beta1 Power accumulator
        """
        assert isinstance(block, framework.Block)
2004
        for param, grad in parameters_and_grads:
C
chengduo 已提交
2005
            if grad is None or param.trainable is False:
2006
                continue
X
Xin Pan 已提交
2007 2008
            with param.block.program._optimized_guard(
                [param, grad]), name_scope('adamx'):
2009 2010
                beta1_pow_acc = self._get_accumulator(self._beta1_pow_acc_str,
                                                      param)
2011
                block.append_op(
2012 2013 2014
                    type="scale",
                    inputs={"X": beta1_pow_acc},
                    outputs={"Out": beta1_pow_acc},
M
minqiyang 已提交
2015 2016
                    attrs={"scale": self._beta1},
                    stop_gradient=True)
2017 2018


2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056
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.
2057 2058 2059
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
2060 2061 2062 2063 2064 2065 2066 2067
    Notes:
       Currently, DpsgdOptimizer doesn't support sparse parameter optimization.
    """

    def __init__(self,
                 learning_rate=0.001,
                 clip=0.9,
                 batch_size=0.999,
2068 2069
                 sigma=1e-8,
                 parameter_list=None):
2070 2071 2072 2073
        assert learning_rate is not None
        assert clip is not None
        assert batch_size is not None
        assert sigma is not None
2074 2075
        super(DpsgdOptimizer, self).__init__(
            learning_rate=learning_rate, parameter_list=parameter_list)
2076 2077 2078 2079
        self.type = "dpsgd"
        self._clip = clip
        self._batch_size = batch_size
        self._sigma = sigma
Z
zhongpu 已提交
2080 2081 2082 2083 2084 2085 2086
        '''
        Note(wangzhongpu):
        This property is only used for debugging, do not need to set it!
        Dpsgd operator use time(NULL) as random seed to generate random number.
        However, during debugging, we need determinated result, so we will set self._seed to a fixed number.
        '''
        self._seed = None
2087 2088 2089 2090 2091

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

        # create the dpsgd optimize op
Z
zhongpu 已提交
2092 2093 2094
        if self._seed == None:
            self._seed = 0

2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105
        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,
Z
zhongpu 已提交
2106 2107
                "sigma": self._sigma,
                "seed": self._seed
2108 2109 2110 2111 2112 2113
            },
            stop_gradient=True)

        return dpsgd_op


2114
class DecayedAdagradOptimizer(Optimizer):
2115
    """
2116 2117 2118
    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.
2119

2120
    The parameter ``param_out`` update rule with gradient ``grad``:
2121 2122 2123 2124 2125 2126 2127

    .. math::

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

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

2128 2129 2130 2131
    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
2132 2133 2134
    stability to avoid the division by zero error.

    Args:
2135 2136 2137 2138 2139
        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.
2140 2141 2142
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
2143 2144 2145 2146 2147 2148 2149 2150
        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.**
2151 2152 2153 2154

    Examples:
        .. code-block:: python

2155 2156
            import paddle.fluid as fluid

2157 2158 2159 2160
            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)
2161
            optimizer.minimize(cost)
2162 2163 2164
    """
    _moment_acc_str = "moment"

X
Xin Pan 已提交
2165 2166 2167 2168
    def __init__(self,
                 learning_rate,
                 decay=0.95,
                 epsilon=1.0e-6,
2169
                 parameter_list=None,
X
Xin Pan 已提交
2170 2171
                 regularization=None,
                 name=None):
2172 2173 2174 2175
        assert learning_rate is not None
        assert decay is not None
        assert epsilon is not None

Q
Qiao Longfei 已提交
2176
        super(DecayedAdagradOptimizer, self).__init__(
X
Xin Pan 已提交
2177
            learning_rate=learning_rate,
2178
            parameter_list=parameter_list,
X
Xin Pan 已提交
2179 2180
            regularization=regularization,
            name=name)
2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207
        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},
2208 2209
            attrs={"epsilon": self._epsilon,
                   "decay": self._decay},
M
minqiyang 已提交
2210
            stop_gradient=True)
2211 2212

        return decayed_adagrad_op
2213 2214


2215
class AdadeltaOptimizer(Optimizer):
2216
    """
Z
Zeng Jinle 已提交
2217
    **Notes: This API does not support sparse parameter optimization.**
Q
qiaolongfei 已提交
2218

Z
Zeng Jinle 已提交
2219
    Adadelta Optimizer. Please refer to this for details:
Z
Zeng Jinle 已提交
2220 2221 2222
    `ADADELTA: AN ADAPTIVE LEARNING RATE METHOD <https://arxiv.org/abs/1212.5701>`_.

    The update is done as follows:
2223

Z
Zeng Jinle 已提交
2224 2225
    .. math::

Z
Zeng Jinle 已提交
2226
        E(g_t^2) &= \\rho * E(g_{t-1}^2) + (1-\\rho) * g^2
2227

Z
Zeng Jinle 已提交
2228
        learning\_rate &= \sqrt{ ( E(dx_{t-1}^2) + \\epsilon ) / ( E(g_t^2) + \\epsilon ) }
Z
Zeng Jinle 已提交
2229

Z
Zeng Jinle 已提交
2230
        E(dx_t^2) &= \\rho * E(dx_{t-1}^2) + (1-\\rho) * (-g*learning\_rate)^2
2231 2232

    Args:
Z
Zeng Jinle 已提交
2233 2234 2235
        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.
2236 2237 2238
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
Z
Zeng Jinle 已提交
2239 2240 2241
        regularization (WeightDecayRegularizer, optional): A Regularizer, such as
                fluid.regularizer.L2DecayRegularizer. Default None, meaning that there is no
                regularization.
2242 2243 2244
        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` .
2245 2246 2247 2248

    Examples:
        .. code-block:: python

2249
            import paddle.fluid as fluid
Z
Zeng Jinle 已提交
2250

2251
            image = fluid.data(name='image', shape=[None, 28], dtype='float32')
Z
Zeng Jinle 已提交
2252 2253
            fc = fluid.layers.fc(image, size=10)
            cost = fluid.layers.reduce_mean(fc)
2254 2255
            optimizer = fluid.optimizer.Adadelta(
                learning_rate=0.0003, epsilon=1.0e-6, rho=0.95)
C
chengduo 已提交
2256

Z
Zeng Jinle 已提交
2257 2258 2259 2260
            # 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)
2261
    """
2262

2263 2264 2265
    _avg_squared_grad_acc_str = "_avg_squared_grad"
    _avg_squared_update_acc_str = "_avg_squared_update"

X
Xin Pan 已提交
2266 2267 2268 2269
    def __init__(self,
                 learning_rate,
                 epsilon=1.0e-6,
                 rho=0.95,
2270
                 parameter_list=None,
X
Xin Pan 已提交
2271 2272
                 regularization=None,
                 name=None):
2273 2274 2275 2276 2277 2278
        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.")
2279
        super(AdadeltaOptimizer, self).__init__(
X
Xin Pan 已提交
2280
            learning_rate=learning_rate,
2281
            parameter_list=parameter_list,
X
Xin Pan 已提交
2282 2283
            regularization=regularization,
            name=name)
2284 2285 2286 2287 2288
        self.type = "adadelta"
        self._epsilon = epsilon
        self._rho = rho

    def _create_accumulators(self, block, parameters):
2289 2290
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")
2291 2292 2293 2294 2295 2296

        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):
2297 2298
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")
2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319

        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
2320 2321
                   "rho": self._rho},
            stop_gradient=True)
2322 2323 2324 2325

        return adadelta_op


Q
qingqing01 已提交
2326 2327 2328 2329 2330 2331 2332 2333 2334 2335
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
2336
        r(w, t) & = \\rho r(w, t-1) + (1 - \\rho)(\\nabla Q_{i}(w))^2
Q
qingqing01 已提交
2337 2338 2339 2340

        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
2341
    each weight. Then dividing the gradient by :math:`sqrt{v(w,t)}`.
Q
qingqing01 已提交
2342 2343 2344 2345 2346 2347

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

    ..  math::

Q
qiaolongfei 已提交
2348
        r(w, t) & = \\rho r(w, t-1) + (1 - \\rho)(\\nabla Q_{i}(w))^2
Q
qingqing01 已提交
2349

2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363
        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
2364 2365 2366 2367
            \\epsilon}} \\nabla Q_{i}(w)

        w & = w - v(w, t)

Q
qiaolongfei 已提交
2368
    where, :math:`\\rho` is a hyperparameter and typical values are 0.9, 0.95
Q
qingqing01 已提交
2369 2370 2371 2372 2373
    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.


2374 2375 2376
    Parameters:
        learning_rate(float): Global learning rate.
        rho(float): rho is :math: `\\rho` in equation, default is 0.95.
Q
qingqing01 已提交
2377
        epsilon(float): :math: `\\epsilon` in equation is smoothing term to
2378
            avoid division by zero, default is 1e-6.
Q
qiaolongfei 已提交
2379
        momentum(float): :math:`\\beta` in equation is the momentum term,
2380
            default is 0.0.
2381 2382 2383 2384
        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.
2385 2386 2387
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
2388 2389 2390 2391
        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
2392 2393 2394 2395 2396 2397 2398

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

    Examples:
          .. code-block:: python

2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423
            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
2424 2425 2426 2427
    """

    _momentum_acc_str = "momentum"
    _mean_square_acc_str = "mean_square"
2428
    _mean_grad_acc_str = "mean_grad"
Q
qingqing01 已提交
2429 2430 2431 2432 2433 2434

    def __init__(self,
                 learning_rate,
                 rho=0.95,
                 epsilon=1.0e-6,
                 momentum=0.0,
2435
                 centered=False,
2436
                 parameter_list=None,
X
Xin Pan 已提交
2437 2438
                 regularization=None,
                 name=None):
Q
qingqing01 已提交
2439
        super(RMSPropOptimizer, self).__init__(
X
Xin Pan 已提交
2440
            learning_rate=learning_rate,
2441
            parameter_list=parameter_list,
X
Xin Pan 已提交
2442 2443
            regularization=regularization,
            name=name)
Q
qingqing01 已提交
2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456
        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
2457
        self._centered = centered
Q
qingqing01 已提交
2458 2459 2460 2461 2462 2463 2464 2465

    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)
2466
            self._add_accumulator(self._mean_grad_acc_str, p)
Q
qingqing01 已提交
2467 2468 2469 2470 2471 2472 2473 2474 2475

    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])
2476 2477
        mean_grad_acc = self._get_accumulator(self._mean_grad_acc_str,
                                              param_and_grad[0])
Q
qingqing01 已提交
2478 2479 2480 2481 2482 2483 2484
        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,
2485
                "MeanGrad": mean_grad_acc,
Q
qingqing01 已提交
2486 2487 2488 2489 2490
                "LearningRate": self._create_param_lr(param_and_grad),
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "MomentOut": momentum_acc,
2491 2492
                "MeanSquareOut": mean_square_acc,
                "MeanGradOut": mean_grad_acc
Q
qingqing01 已提交
2493 2494 2495 2496
            },
            attrs={
                "epsilon": self._epsilon,
                "decay": self._rho,
2497 2498
                "momentum": self._momentum,
                "centered": self._centered
M
minqiyang 已提交
2499 2500
            },
            stop_gradient=True)
Q
qingqing01 已提交
2501 2502 2503 2504

        return rmsprop_op


Q
qiaolongfei 已提交
2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544
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

2545 2546 2547 2548 2549
    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.
2550 2551 2552
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
2553 2554 2555 2556
        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
2557 2558 2559 2560 2561 2562 2563

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

    Examples:
          .. code-block:: python

2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587
            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
2588

2589
    NOTE:
C
chengduo 已提交
2590
       Currently, FtrlOptimizer doesn't support sparse parameter optimization.
Q
qiaolongfei 已提交
2591 2592 2593 2594 2595
    """

    _squared_acc_str = "squared"
    _linear_acc_str = "linear"

X
Xin Pan 已提交
2596 2597 2598 2599 2600
    def __init__(self,
                 learning_rate,
                 l1=0.0,
                 l2=0.0,
                 lr_power=-0.5,
2601
                 parameter_list=None,
X
Xin Pan 已提交
2602 2603
                 regularization=None,
                 name=None):
Q
qiaolongfei 已提交
2604
        super(FtrlOptimizer, self).__init__(
X
Xin Pan 已提交
2605
            learning_rate=learning_rate,
2606
            parameter_list=parameter_list,
X
Xin Pan 已提交
2607 2608
            regularization=regularization,
            name=name)
Q
qiaolongfei 已提交
2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648
        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
2649 2650
                   "lr_power": self._lr_power},
            stop_gradient=True)
Q
qiaolongfei 已提交
2651 2652 2653 2654

        return ftrl_op


Y
Yibing Liu 已提交
2655 2656 2657 2658 2659 2660
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
2661 2662
    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
2663 2664 2665 2666 2667

    The updating of parameters follows:

    ..  math::

Y
Yibing Liu 已提交
2668
        m_t &= \\beta_1 m_{t - 1}+ (1 - \\beta_1)g_t 
Y
Yibing Liu 已提交
2669

Y
Yibing Liu 已提交
2670
        v_t &= \\beta_2 v_{t - 1}  + (1 - \\beta_2)g_t^2
Y
Yibing Liu 已提交
2671

Y
Yibing Liu 已提交
2672
        r_t &= \\frac{m_t}{\\sqrt{v_t}+\\epsilon}
Y
Yibing Liu 已提交
2673

Y
Yibing Liu 已提交
2674
        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
2675 2676 2677 2678 2679 2680


    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
2681 2682 2683 2684 2685 2686 2687 2688
        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.
2689 2690 2691
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
Y
Yibing Liu 已提交
2692 2693 2694 2695 2696 2697 2698
        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
2699 2700 2701 2702 2703 2704

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

Y
Yibing Liu 已提交
2705
            data = fluid.data(name='x', shape=[-1, 5], dtype='float32')
Y
Yibing Liu 已提交
2706 2707 2708
            hidden = fluid.layers.fc(input=data, size=10)
            cost = fluid.layers.mean(hidden)

Y
Yibing Liu 已提交
2709 2710 2711 2712 2713
            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
2714 2715 2716 2717
            optimizer.minimize(cost)
    """
    _moment1_acc_str = "moment1"
    _moment2_acc_str = "moment2"
Y
Yibing Liu 已提交
2718
    # these two not used in op temporarily
Y
Yibing Liu 已提交
2719 2720 2721 2722 2723 2724 2725 2726 2727
    _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,
2728
                 parameter_list=None,
Y
Yibing Liu 已提交
2729
                 regularization=None,
Y
Yibing Liu 已提交
2730
                 exclude_from_weight_decay_fn=None,
Y
Yibing Liu 已提交
2731 2732 2733 2734 2735 2736 2737 2738
                 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,
2739
            parameter_list=parameter_list,
Y
Yibing Liu 已提交
2740 2741 2742 2743 2744 2745 2746
            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
2747
        self._exclude_from_weight_decay_fn = exclude_from_weight_decay_fn
Y
Yibing Liu 已提交
2748 2749 2750

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)
2751
        block.program._use_lamb = True
Y
Yibing Liu 已提交
2752 2753 2754 2755 2756 2757 2758 2759 2760 2761

        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
2762 2763 2764 2765 2766 2767
        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
2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788
        # 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
2789
                "weight_decay": weight_decay
Y
Yibing Liu 已提交
2790 2791 2792 2793 2794 2795
            },
            stop_gradient=True)

        return lamb_op


2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808
# 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
2809
Dpsgd = DpsgdOptimizer
2810
DecayedAdagrad = DecayedAdagradOptimizer
2811
Adadelta = AdadeltaOptimizer
Q
qingqing01 已提交
2812
RMSProp = RMSPropOptimizer
Q
qiaolongfei 已提交
2813
Ftrl = FtrlOptimizer
2814
LarsMomentum = LarsMomentumOptimizer
Y
Yibing Liu 已提交
2815
Lamb = LambOptimizer
2816 2817 2818


class ModelAverage(Optimizer):
2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837
    """
    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:

    ::
2838

2839 2840 2841 2842 2843 2844 2845 2846 2847
        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.
2848 2849

    Args:
2850 2851 2852 2853 2854 2855 2856 2857
        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.
2858

2859
    Examples:
Q
qiaolongfei 已提交
2860 2861 2862

      .. code-block:: python

2863 2864 2865 2866 2867 2868
        import paddle.fluid as fluid
        import numpy

        # First create the Executor.
        place = fluid.CPUPlace()  # fluid.CUDAPlace(0)
        exe = fluid.Executor(place)
2869

2870 2871 2872 2873
        train_program = fluid.Program()
        startup_program = fluid.Program()
        with fluid.program_guard(train_program, startup_program):
            # build net
2874
            data = fluid.data(name='X', shape=[None, 1], dtype='float32')
2875 2876 2877 2878 2879 2880 2881 2882
            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,
2883
                                                         max_average_window=12500)
2884 2885

            exe.run(startup_program)
2886 2887 2888 2889 2890
            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])
2891 2892

            # apply ModelAverage
2893
            with model_average.apply(exe):
2894 2895 2896 2897
                x = numpy.random.random(size=(10, 1)).astype('float32')
                exe.run(program=train_program,
                        feed={'X': x},
                        fetch_list=[loss.name])
2898 2899 2900
    """

    def __init__(self,
W
wanghaoshuang 已提交
2901
                 average_window_rate,
2902 2903
                 min_average_window=10000,
                 max_average_window=10000,
X
Xin Pan 已提交
2904 2905
                 regularization=None,
                 name=None):
Z
zhongpu 已提交
2906 2907
        if framework.in_dygraph_mode():
            raise Exception("In dygraph, don't support ModelAverage.")
X
Xin Pan 已提交
2908 2909
        super(ModelAverage, self).__init__(
            0.0, regularization=regularization, name=name)
2910 2911 2912
        self.average_window = average_window_rate
        self.min_average_window = min_average_window
        self.max_average_window = max_average_window
2913

2914
        self.params_grads = []
2915 2916
        for param in framework.default_main_program().global_block(
        ).all_parameters():
2917
            if param.do_model_average != False:
2918
                grad = param.block.create_var(
2919 2920
                    name=unique_name.generate_with_ignorable_key(".".join(
                        [param.name, 'tmp'])),
2921 2922
                    dtype=param.dtype,
                    persistable=False,
W
wanghaoshuang 已提交
2923
                    stop_gradient=True)
2924
                self.params_grads.append((param, grad))
2925

2926
        for param, grad in self.params_grads:
2927 2928
            if grad is None:
                continue
X
Xin Pan 已提交
2929 2930
            with param.block.program._optimized_guard(
                [param, grad]), name_scope('move_average'):
2931
                self._append_average_accumulate_op(param)
2932

2933 2934 2935 2936
        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:
2937
                self._add_average_apply_op(block, param_grad)
2938 2939 2940 2941 2942

        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:
2943
                self._add_average_restore_op(block, param_grad)
2944

2945
    def _add_average_apply_op(self, block, param_grad):
L
Luo Tao 已提交
2946 2947 2948 2949 2950 2951
        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(
2952
            self._get_accumulator('num_accumulates', param))
L
Luo Tao 已提交
2953
        old_num_accumulates = block._clone_variable(
2954
            self._get_accumulator('old_num_accumulates', param))
L
Luo Tao 已提交
2955
        num_updates = block._clone_variable(
2956 2957 2958 2959 2960 2961
            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
2962 2963 2964 2965
        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
2966
        ops._elementwise_div(x=sum, y=tmp, out=param)
2967 2968

    def _add_average_restore_op(self, block, param_grad):
L
Luo Tao 已提交
2969 2970
        param = block._clone_variable(param_grad[0])
        grad = block._clone_variable(param_grad[1])
2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007
        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
3008 3009
            },
            stop_gradient=True)
3010

S
rename  
sneaxiy 已提交
3011
    @signature_safe_contextmanager
3012
    def apply(self, executor, need_restore=True):
3013 3014
        """
        Apply the average of the cumulative ``Parameter`` to the parameters of the current model.
3015 3016

        Args:
3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060
            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])
3061
        """
3062 3063 3064 3065 3066 3067
        executor.run(self.apply_program)
        try:
            yield
        finally:
            if need_restore:
                self.restore(executor)
3068 3069

    def restore(self, executor):
3070 3071
        """
        Restore ``Parameter`` values of current model.
3072 3073
        
        Args:
3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117
            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)
3118
        """
3119
        executor.run(self.restore_program)
3120 3121 3122 3123 3124 3125 3126 3127 3128 3129


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::

3130
        \\text{EMA}_0 & = 0
3131

3132 3133
	\\text{EMA}_t & = \\text{decay} * \\text{EMA}_{t-1} + (1 - \\text{decay}) * \\theta_t

Y
Yibing Liu 已提交
3134 3135 3136 3137
    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.
3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158

    **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.
3159 3160 3161


    Args:
Y
Yibing Liu 已提交
3162 3163 3164 3165 3166 3167 3168
	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.
3169 3170 3171 3172 3173


    Examples:

	.. code-block:: python
3174 3175 3176 3177 3178

	    import numpy
	    import paddle
	    import paddle.fluid as fluid

Y
Yibing Liu 已提交
3179
	    data = fluid.data(name='x', shape=[-1, 5], dtype='float32')
3180 3181 3182 3183 3184 3185 3186 3187
	    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)

3188
	    global_steps = fluid.layers.autoincreased_step_counter()
3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217
	    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)
3218 3219
    """

3220
    def __init__(self, decay=0.999, thres_steps=None, name=None):
Z
zhongpu 已提交
3221 3222 3223
        if framework.in_dygraph_mode():
            raise Exception(
                "In dygraph, don't support ExponentialMovingAverage.")
3224
        self._decay = decay
3225
        self._thres_steps = thres_steps
3226
        self._name = name if name is not None else ''
3227 3228
        self._decay_var = self._get_ema_decay()

3229
        self._step_counter_name = "@EMA_STEP_COUNTER@"
Y
Yibing Liu 已提交
3230
        self._params_tmps = []
3231
        for param in default_main_program().global_block().all_parameters():
3232 3233 3234 3235 3236 3237 3238
            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
3239
                self._params_tmps.append((param, tmp))
3240

Y
Yibing Liu 已提交
3241 3242
        self._ema_vars = {}
        for param, tmp in self._params_tmps:
3243 3244
            with param.block.program._optimized_guard(
                [param, tmp]), name_scope('moving_average'):
Y
Yibing Liu 已提交
3245
                self._ema_vars[param.name] = self._create_ema_vars(param)
3246 3247 3248 3249

        self.apply_program = Program()
        block = self.apply_program.global_block()
        with program_guard(main_program=self.apply_program):
3250
            decay_pow, global_step = self._get_decay_pow(block)
Y
Yibing Liu 已提交
3251
            for param, tmp in self._params_tmps:
3252 3253
                param = block._clone_variable(param)
                tmp = block._clone_variable(tmp)
Y
Yibing Liu 已提交
3254
                ema = block._clone_variable(self._ema_vars[param.name])
3255
                layers.assign(input=param, output=tmp)
3256
                # bias correction
3257 3258 3259
                with layers.control_flow.Switch() as switch:
                    with switch.case(global_step > 0):
                        layers.assign(output=ema, input=ema / (1.0 - decay_pow))
3260 3261 3262 3263 3264
                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
3265
            for param, tmp in self._params_tmps:
3266 3267 3268 3269
                tmp = block._clone_variable(tmp)
                param = block._clone_variable(param)
                layers.assign(input=tmp, output=param)

3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291
    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):
3292 3293 3294 3295 3296 3297 3298
        global_step = layers.create_global_var(
            name=self._step_counter_name,
            shape=[1],
            value=0,
            dtype='int64',
            persistable=True)
        global_step = layers.cast(global_step, "float32")
3299
        decay_var = block._clone_variable(self._decay_var)
3300 3301
        decay_pow_acc = layers.elementwise_pow(decay_var, global_step)
        return decay_pow_acc, global_step
3302

Y
Yibing Liu 已提交
3303
    def _create_ema_vars(self, param):
3304 3305 3306 3307 3308 3309 3310 3311 3312
        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
3313 3314 3315 3316 3317
    def update(self):
        """ 
        Update Exponential Moving Average. Should only call this method in 
        train program.
        """
3318 3319
        global_step = layers.autoincreased_step_counter(
            counter_name=self._step_counter_name)
3320
        param_master_emas = []
Y
Yibing Liu 已提交
3321 3322 3323 3324
        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]
3325
                if param.name + '.master' in self._ema_vars:
3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342
                    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
3343

3344 3345 3346 3347 3348 3349 3350
    @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
3351 3352
            need_restore (bool, optional): Whether to restore parameters after 
                applying. Default True.
3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367
        """
        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
3368 3369 3370


class PipelineOptimizer(object):
3371 3372
    """
    Pipeline Optimizer
H
hutuxian 已提交
3373 3374 3375 3376 3377 3378 3379 3380 3381

    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, \
3382
    the final performance depends on the training progress of each pipeline heavily.
H
hutuxian 已提交
3383 3384 3385

    And we will try the synchronous mode in the future.

3386
    Args:
H
hutuxian 已提交
3387 3388 3389 3390
        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.
3391 3392
        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
3393 3394 3395 3396
                        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].

3397 3398
    Examples:
        .. code-block:: python
H
hutuxian 已提交
3399

3400
            import paddle.fluid as fluid
H
hutuxian 已提交
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
            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)
3435 3436
    """

H
hutuxian 已提交
3437 3438 3439 3440 3441 3442 3443 3444
    def __init__(self,
                 optimizer,
                 cut_list=None,
                 place_list=None,
                 concurrency_list=None,
                 queue_size=30,
                 sync_steps=1,
                 start_cpu_core_id=0):
Z
zhongpu 已提交
3445 3446
        if framework.in_dygraph_mode():
            raise Exception("In dygraph, don't support PipelineOptimizer.")
H
hutuxian 已提交
3447 3448 3449 3450 3451 3452 3453 3454 3455
        # 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
3456
    def _create_vars(self, block, main_program):
H
hutuxian 已提交
3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467
        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
3468
    def _extract_section_opt_ops(self, ops, cut_point_name):
H
hutuxian 已提交
3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483
        """
        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
3484
    def _find_input_output(self, ops, name, is_forward=True):
H
hutuxian 已提交
3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498
        """
        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
3499
    def _find_persistable_vars(self, ops, whole_parameters):
H
hutuxian 已提交
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
        """
        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
3527
    def _extract_section_ops(self, ops, cut_point_name):
H
hutuxian 已提交
3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546
        """
        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
3547 3548
    def _find_section_opt(self, ops, params):
        res = self._extract_section_opt_ops(ops, params)
H
hutuxian 已提交
3549 3550
        return res

H
hutuxian 已提交
3551
    def _split_program(self, main_program, cut_list):
H
hutuxian 已提交
3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571
        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
3572
            cur_ops = self._extract_section_ops(ops, cut_vars)
H
hutuxian 已提交
3573 3574 3575 3576 3577 3578
            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
3579
                self._find_input_output(
H
hutuxian 已提交
3580 3581 3582 3583 3584 3585
                    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
3586
                    self._find_persistable_vars(cur_ops, whole_parameters))
H
hutuxian 已提交
3587
            if i >= cut_len - 1:
H
hutuxian 已提交
3588 3589
                opt_ops = self._find_section_opt(
                    ops, sec_params[2 * cut_len - 2 - i])
H
hutuxian 已提交
3590 3591 3592 3593 3594 3595 3596 3597 3598 3599

                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
3600
                self._find_input_output(
H
hutuxian 已提交
3601 3602 3603
                    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
3604
                self._find_input_output(
H
hutuxian 已提交
3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618
                    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
3619
            self._find_input_output(
H
hutuxian 已提交
3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639
                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
3640 3641 3642 3643 3644 3645 3646 3647
        if len(self._cut_list) == 0:
            program_list = []
            ptmp = {"program": program, "input_set": set(), "output_set": set()}
            program_list.append(ptmp)
        else:
            program_list = self._split_program(program, self._cut_list)
            for p in program_list:
                self._create_vars(p["program"].block(0), program)
H
hutuxian 已提交
3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667
        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
3668 3669


M
mapingshuo 已提交
3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 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
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):
Z
zhongpu 已提交
3732 3733
        if framework.in_dygraph_mode():
            raise Exception("In dygraph, don't support RecomputeOptimizer.")
M
mapingshuo 已提交
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
        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`.
3839 3840
            parameter_list (list): list of Variables or Variable.names to update.
            no_grad_set (set|None): set of Variables or Variables.names should be ignored.
M
mapingshuo 已提交
3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899
            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
3900

M
mapingshuo 已提交
3901 3902 3903 3904 3905 3906 3907
                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
3908
                    return sum_cost, fc_1, prediction                
M
mapingshuo 已提交
3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927
                
                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
3928

M
mapingshuo 已提交
3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964
        """

        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
3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019
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):

Z
zhongpu 已提交
4020 4021
        if framework.in_dygraph_mode():
            raise Exception("In dygraph, don't support LookaheadOptimizer.")
M
mapingshuo 已提交
4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113 4114 4115 4116 4117
        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