optimizer.py 73.7 KB
Newer Older
1
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
D
dzhwinter 已提交
2
#
D
dzhwinter 已提交
3 4 5
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
D
dzhwinter 已提交
6
#
D
dzhwinter 已提交
7
#     http://www.apache.org/licenses/LICENSE-2.0
D
dzhwinter 已提交
8
#
D
dzhwinter 已提交
9 10 11 12 13
# 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
from collections import defaultdict
S
rename  
sneaxiy 已提交
18
from .wrapped_decorator import signature_safe_contextmanager
19

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

23 24
from . import framework
from . import layers
25
from . import unique_name
26
from .backward import append_backward
27
from .clip import append_gradient_clip_ops, error_clip_callback
28 29 30
from .framework import program_guard
from .initializer import Constant
from .layer_helper import LayerHelper
S
sneaxiy 已提交
31
from .layers import ops
32
from .regularizer import append_regularization_ops
33 34
from .dygraph import base as imperative_base
from .dygraph.learning_rate_scheduler import LearningRateDecay
35 36 37 38
from paddle.fluid import core
from paddle.fluid.layers import tensor
from functools import reduce
import copy
39

40
__all__ = [
Q
qiaolongfei 已提交
41
    'SGD', 'Momentum', 'Adagrad', 'Adam', 'Adamax', 'DecayedAdagrad', 'Ftrl',
42
    'SGDOptimizer', 'MomentumOptimizer', 'AdagradOptimizer', 'AdamOptimizer',
W
weixing02 已提交
43
    'AdamaxOptimizer', 'DecayedAdagradOptimizer', 'RMSPropOptimizer',
44
    'FtrlOptimizer', 'Adadelta', 'ModelAverage', 'LarsMomentum',
45
    'LarsMomentumOptimizer', 'DGCMomentumOptimizer'
46
]
Q
Qiao Longfei 已提交
47 48 49 50 51 52


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

    Define the common interface of an optimizer.
53 54
    User should not use this class directly,
    but need to use one of it's implementation.
Q
Qiao Longfei 已提交
55 56
    """

X
Xin Pan 已提交
57
    def __init__(self, learning_rate, regularization=None, name=None):
L
lujun 已提交
58
        if framework.in_dygraph_mode():
M
minqiyang 已提交
59 60 61 62 63 64 65 66 67 68 69 70
            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))
        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))

W
whs 已提交
71
        self._name = name
D
dzhwinter 已提交
72
        self.regularization = regularization
73
        self._learning_rate = learning_rate
D
dzhwinter 已提交
74 75
        # the learning rate type should be inferenced from loss
        self._dtype = None
76
        # each program should have a independent learning rate
77
        # program -> Variable(learning_rate)
Q
qiaolongfei 已提交
78
        self._learning_rate_map = dict()
79
        if isinstance(self._learning_rate, framework.Variable):
80 81
            self._learning_rate_map[framework.default_main_program(
            )] = self._learning_rate
82 83 84 85 86
        # 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
87
        self.helper = None
88 89 90 91
        self._opti_name_list = []

    def get_opti_var_name_list(self):
        return self._opti_name_list
Q
Qiao Longfei 已提交
92

Q
Qiao Longfei 已提交
93
    def _create_global_learning_rate(self):
94 95 96
        if imperative_base.enabled():
            # create learning rate Variable
            if isinstance(self._learning_rate, float):
M
minqiyang 已提交
97 98 99 100 101 102 103 104 105 106 107 108
                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)
109
            # get learning rate Variable from LearningRateDecay
M
minqiyang 已提交
110
            elif isinstance(self._learning_rate, LearningRateDecay):
111 112 113
                self._learning_rate_map[framework.default_main_program(
                )] = self._learning_rate()
            else:
Q
qiaolongfei 已提交
114
                raise TypeError(
115 116
                    "optimizer's learning rate must be float or LearningRateDecay"
                )
117
        else:
118 119 120 121
            lr = self._global_learning_rate()

            if isinstance(lr, framework.Variable):
                return
M
minqiyang 已提交
122 123 124 125 126 127
            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
128

129 130 131 132 133 134 135 136
            # 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)
137

Y
yuyang18 已提交
138
    def _global_learning_rate(self, program=None):
Q
Qiao Longfei 已提交
139 140 141 142
        """
        get global decayed learning rate
        :return:
        """
143 144
        if program is None:
            program = framework.default_main_program()
Q
qiaolongfei 已提交
145
        return self._learning_rate_map.get(program, None)
Q
Qiao Longfei 已提交
146

Q
Qiao Longfei 已提交
147 148 149 150 151
    def _append_optimize_op(self, block, param_and_grad):
        """ append optimize operator to block and return all the added optimize_op
        """
        raise NotImplementedError()

152 153 154 155
    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
156 157
        if type(param_lr) == Variable:
            return param_lr
Q
qiaolongfei 已提交
158
        else:
W
Wu Yi 已提交
159
            if param_lr == 1.0:
Y
yuyang18 已提交
160
                return self._global_learning_rate()
W
Wu Yi 已提交
161
            else:
X
Xin Pan 已提交
162 163 164
                with default_main_program()._lr_schedule_guard(
                        is_with_opt=True), framework.name_scope(
                            'scale_with_param_lr'):
165
                    return self._global_learning_rate() * param_lr
166 167 168 169 170 171 172

    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
173
        """
174 175
        pass

176
    def _finish_update(self, block, parameters_and_grads):
177 178 179 180 181 182 183 184
        """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
185
            None
186 187 188
        """
        pass

189 190 191 192 193 194
    def _add_accumulator(self,
                         name,
                         param,
                         dtype=None,
                         fill_value=0.0,
                         shape=None):
195 196 197 198 199 200 201 202 203
        """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
204 205
        if self._name is not None:
            name = self._name + "_" + name
206 207
        if (name in self._accumulators and
                param.name in self._accumulators[name]):
L
lujun 已提交
208
            if framework.in_dygraph_mode():
X
polish  
Xin Pan 已提交
209
                return self._accumulators[name][param.name]
210
            raise Exception("Accumulator {} already exists for parameter {}".
211
                            format(name, param.name))
212 213
        if shape == None:
            shape = param.shape
Q
Qiao Longfei 已提交
214
        assert isinstance(self.helper, LayerHelper)
215 216 217 218 219

        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
220
        var = self.helper.create_global_variable(
221
            name=var_name,
Q
Qiao Longfei 已提交
222
            persistable=True,
F
fengjiayi 已提交
223
            dtype=dtype or param.dtype,
Q
Qiao Longfei 已提交
224
            type=param.type,
225
            shape=shape)
Q
Qiao Longfei 已提交
226
        self.helper.set_variable_initializer(
227
            var, initializer=Constant(value=float(fill_value)))
Q
Qiao Longfei 已提交
228
        self._accumulators[name][param.name] = var
229
        return var
230 231 232 233 234 235 236 237 238 239 240

    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
241 242
        if self._name is not None:
            name = self._name + "_" + name
243 244 245 246 247 248
        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]

249
    def _create_optimization_pass(self, parameters_and_grads):
Q
Qiao Longfei 已提交
250 251 252
        """Add optimization operators to update gradients to variables.

        Args:
Q
qiaolongfei 已提交
253
          parameters_and_grads(list(tuple(Variable, Variable))):
254
            a list of (variable, gradient) pair to update.
Q
Qiao Longfei 已提交
255 256

        Returns:
257
          return_op_list: a list of operators that will complete one step of
258 259 260
            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
261
        """
262 263 264 265 266
        # 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
267
        # for parameters and extend _finish_update method to add custom ops.
268

269 270 271 272 273 274 275 276 277
        # Allways called under program_guard use global block as loss block
        global_block = framework.default_main_program().global_block()
        start = len(global_block.ops)
        self.helper = LayerHelper(self.__class__.__name__)
        self._create_accumulators(global_block,
                                  [p[0] for p in parameters_and_grads])
        self._create_global_learning_rate()

        optimize_ops = []
278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297
        if framework._in_dygraph_mode():
            for param_and_grad in parameters_and_grads:
                if param_and_grad[1] is None:
                    continue
                with param_and_grad[0].block.program._optimized_guard(
                        param_and_grad):
                    if param_and_grad[0].trainable is True:
                        optimize_op = self._append_optimize_op(global_block,
                                                               param_and_grad)
                        optimize_ops.append(optimize_op)
        else:
            for param_and_grad in parameters_and_grads:
                if param_and_grad[1] is None:
                    continue
                with param_and_grad[0].block.program._optimized_guard(
                        param_and_grad), name_scope("optimizer"):
                    if param_and_grad[0].trainable is True:
                        optimize_op = self._append_optimize_op(global_block,
                                                               param_and_grad)
                        optimize_ops.append(optimize_op)
298 299 300 301 302 303 304 305 306

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

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

    def _process_distribute_lookuptable(self, param_grads):
Q
Qiao Longfei 已提交
307 308 309 310 311 312 313 314 315
        """
        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
        """
316 317
        program = framework.default_main_program()
        global_block = framework.default_main_program().global_block()
Q
Qiao Longfei 已提交
318 319 320 321 322 323 324 325 326 327 328 329 330 331 332
        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:
333 334 335 336 337 338 339 340 341 342 343 344 345
            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
346 347
        return new_param_grads, (table_param, table_grad), sgd_op

348 349 350
    def _append_dgc_ops(self, param_and_grad):
        pass

351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368
    def backward(self,
                 loss,
                 startup_program=None,
                 parameter_list=None,
                 no_grad_set=None,
                 callbacks=None):
        """
        First part of `minimize`, do auto-diff to append backward ops for
        the current program.

        Args:
            loss (Variable): loss variable to run optimizations.
            startup_program (Program): startup_program for initializing parameters
                in `parameter_list`.
            parameter_list (list): list of Variables to update.
            no_grad_set (set|None): set of Variables should be ignored.
            callbacks (list|None): list of callables to run when appending backward
                operator for one parameter.
M
minqiyang 已提交
369

370 371
        Return:
            list: list of (param, grad) pair, grad is the output of backward.
M
minqiyang 已提交
372

373 374 375
        Examples:
            See examples in `apply_gradients`.
        """
C
chengduo 已提交
376
        self._dtype = loss.dtype
L
lujun 已提交
377
        if framework.in_dygraph_mode():
C
chengduo 已提交
378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394
            if parameter_list is not None:
                parameters = parameter_list
            else:
                parameters = framework._dygraph_tracer().all_parameters()

            params_grads = []
            for param in parameters:
                if not param.trainable:
                    continue
                if param._ivar._grad_ivar() is not None:
                    # create gradient variable
                    grad_var = Variable(
                        block=loss.block,
                        name=param._ivar._grad_name(),
                        stop_gradient=True,
                        ivar=param._ivar._grad_ivar())
                    params_grads.append((param, grad_var))
395
        else:
C
chengduo 已提交
396 397 398 399 400 401 402 403 404 405 406 407
            if callbacks is None:
                callbacks = [error_clip_callback]
            else:
                assert (isinstance(callbacks, list))
            program = loss.block.program
            with program_guard(program, startup_program):
                params_grads = append_backward(loss, parameter_list,
                                               no_grad_set, callbacks)
                # Note: since we can't use all_reduce_op now,
                #  dgc_op should be the last op of one grad.
                self._append_dgc_ops(params_grads)
        return params_grads
408 409 410 411 412 413 414 415

    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
416

417 418
        Returns:
            list: A list of operators appended to the current program.
M
minqiyang 已提交
419

420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447
        Examples:
            .. code-block:: python

                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
448 449 450 451 452 453 454 455 456 457 458 459 460 461
    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
462
        if framework.in_dygraph_mode():
C
chengduo 已提交
463 464 465 466 467 468 469 470 471
            with program_guard(framework.default_main_program(),
                               framework.default_startup_program()):
                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

Q
Qiao Longfei 已提交
472 473
    def minimize(self,
                 loss,
474
                 startup_program=None,
Q
Qiao Longfei 已提交
475 476
                 parameter_list=None,
                 no_grad_set=None):
477 478 479 480 481
        """
        Add operations to minimize `loss` by updating `parameter_list`.

        This method combines interface `backward()` and
        `apply_gradients()` into one.
M
minqiyang 已提交
482

483 484 485 486 487 488
        Args:
            loss (Variable): loss variable to run optimizations.
            startup_program (Program): startup_program for initializing parameters
                in `parameter_list`.
            parameter_list (list): list of Variables to update.
            no_grad_set (set|None): set of Variables should be ignored.
Q
Qiao Longfei 已提交
489

490 491 492
        Returns:
            tuple: (optimize_ops, params_grads) which are, list of operators appended;
            and list of (param, grad) Variables pair for optimization.
Q
Qiao Longfei 已提交
493
        """
C
chengduo 已提交
494 495 496 497 498 499 500
        params_grads = self.backward(
            loss,
            startup_program=startup_program,
            parameter_list=parameter_list,
            no_grad_set=no_grad_set)
        optimize_ops = self.apply_optimize(
            loss, startup_program=startup_program, params_grads=params_grads)
M
minqiyang 已提交
501

Q
Qiao Longfei 已提交
502
        return optimize_ops, params_grads
Q
Qiao Longfei 已提交
503 504 505


class SGDOptimizer(Optimizer):
Q
qiaolongfei 已提交
506 507 508 509 510 511 512 513 514 515
    """
    Optimizer of the stochastic gradient descent algorithm.

    .. math::

        param\_out = param - learning\_rate * grad

    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.
X
Xin Pan 已提交
516 517 518
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
Q
qiaolongfei 已提交
519 520 521 522

    Examples:
        .. code-block:: python

Q
qiaolongfei 已提交
523
            sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.2)
Q
qiaolongfei 已提交
524
            sgd_optimizer.minimize(cost)
Q
Qiao Longfei 已提交
525 526
    """

X
Xin Pan 已提交
527
    def __init__(self, learning_rate, regularization=None, name=None):
Q
Qiao Longfei 已提交
528
        assert learning_rate is not None
Q
Qiao Longfei 已提交
529
        super(SGDOptimizer, self).__init__(
X
Xin Pan 已提交
530 531 532
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
Q
Qiao Longfei 已提交
533 534
        self.type = "sgd"

535 536
    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)
537

Q
Qiao Longfei 已提交
538 539 540 541 542 543
        # create the optimize op
        sgd_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
544
                "LearningRate": self._create_param_lr(param_and_grad)
Q
Qiao Longfei 已提交
545
            },
M
minqiyang 已提交
546 547
            outputs={"ParamOut": param_and_grad[0]},
            stop_gradient=True)
Q
Qiao Longfei 已提交
548 549

        return sgd_op
550 551 552


class MomentumOptimizer(Optimizer):
Q
qiaolongfei 已提交
553 554 555 556 557 558 559 560 561 562 563 564 565 566
    """

    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):

567
        &\quad   param = param - (gradient + mu * velocity) * learning\_rate
Q
qiaolongfei 已提交
568 569 570

        & else:

Q
qiaolongfei 已提交
571
        &\quad   param = param - learning\_rate * velocity
Q
qiaolongfei 已提交
572 573 574 575 576 577

    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.
        momentum (float): momentum factor
        use_nesterov (bool): enables Nesterov momentum
X
Xin Pan 已提交
578 579 580
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
Q
qiaolongfei 已提交
581 582 583 584

    Examples:
        .. code-block:: python

Q
qiaolongfei 已提交
585
            optimizer = fluid.optimizer.Momentum(learning_rate=0.2, momentum=0.1)
Q
qiaolongfei 已提交
586
            optimizer.minimize(cost)
587 588 589
    """
    _velocity_acc_str = "velocity"

X
Xin Pan 已提交
590 591 592 593 594 595
    def __init__(self,
                 learning_rate,
                 momentum,
                 use_nesterov=False,
                 regularization=None,
                 name=None):
596 597
        assert learning_rate is not None
        assert momentum is not None
Q
Qiao Longfei 已提交
598
        super(MomentumOptimizer, self).__init__(
X
Xin Pan 已提交
599 600 601
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
602 603
        self.type = "momentum"
        self._momentum = momentum
604
        self._use_nesterov = bool(use_nesterov)
605 606 607 608 609

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

        for p in parameters:
Q
Qiao Longfei 已提交
610
            self._add_accumulator(self._velocity_acc_str, p)
611 612 613 614 615 616 617 618 619 620 621 622 623

    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,
624
                "LearningRate": self._create_param_lr(param_and_grad)
625 626 627 628 629
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "VelocityOut": velocity_acc
            },
630
            attrs={"mu": self._momentum,
M
minqiyang 已提交
631 632
                   "use_nesterov": self._use_nesterov},
            stop_gradient=True)
633 634

        return momentum_op
635 636


637 638 639 640 641
class DGCMomentumOptimizer(MomentumOptimizer):
    """

    Original paper is https://arxiv.org/abs/1712.01887

G
gongweibao 已提交
642
    DGC reduces the communication bandwidth by sending only the important gradients (sparse update):\
643 644
        only gradients larger than a threshold are transmitted.

G
gongweibao 已提交
645
    To avoid losing information, DGC accumulates the rest of the gradients locally.
646 647 648

    Eventually, these gradients become large enough to be transmitted.

G
gongweibao 已提交
649
    Thus, DGC sends the large gradients immediately but eventually send all of the gradients over time.
650

G
gongweibao 已提交
651
    To ensure no loss of accuracy, DGC employs momentum correction and local gradient clipping on top of the gradient sparsification to maintain model performance.
652 653 654 655

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

    This optimizer will do two things:
656

657 658
        1. Compress the gradient by get TopK import value from tensor \
            and use it for allreduce to reduce network bandwidth.
659

660 661 662 663 664 665
        2. Call momentum to optimize on the cost.

    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.
        momentum (float): Momentum factor.
G
gongweibao 已提交
666
        rampup_begin_step (int): The beginning step from which gradient compression is implemented.
667 668 669 670 671 672 673
        rampup_step (int): How long it use the sparsity 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 5, \
                it will use 0.75 at 0 step, and 0.9375 at 1 step, 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).
        use_nesterov (bool): Enables Nesterov momentum. True means use nesterov.
        local_grad_clip_norm (float): Clip norm value if needed.
G
gongweibao 已提交
674
        num_trainers: The number of training nodes.
675
        regularization: A Regularizer, such as fluid.regularizer.L2DecayRegularizer.
G
gongweibao 已提交
676
        name: An optional name prefix.
677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765

    Examples:
        .. code-block:: python

            optimizer = fluid.optimizer.DGCMomentumOptimizer(
                learning_rate=fluid.layers.piecewise_decay(
                    boundaries=bd, values=lr),
                momentum=0.9,
                rampup_begin_step=1252,
                regularization=fluid.regularizer.L2Decay(1e-4))
            optimizer.minimize(cost)

    """

    def __init__(self,
                 learning_rate,
                 momentum,
                 rampup_begin_step,
                 rampup_step=1,
                 sparsity=[0.999],
                 use_nesterov=False,
                 local_grad_clip_norm=None,
                 num_trainers=None,
                 regularization=None,
                 name=None):
        self._sparsity = sparsity
        self._rampup_step = rampup_step
        self._rampup_step_var = None

        self._rampup_begin_step = rampup_begin_step
        self._rampup_begin_step_var = None

        self._global_step_var = None
        self._local_grad_clip_norm = None
        self._clip_norm = None

        if local_grad_clip_norm is not None:
            assert isinstance(num_trainers, int)
            assert isinstance(local_grad_clip_norm, float)
            assert num_trainers > 0

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

        super(DGCMomentumOptimizer, self).__init__(
            learning_rate, momentum, use_nesterov, regularization, name)

        core.init_dgc()

    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

    def _append_dgc_ops(self, param_and_grads):
        start_program = default_startup_program()
        main_program = default_main_program()
        main_program._enable_dgc = True

        # step counter
        self._global_step_var = self._add_auto_increment_var(
            counter_name='__g_dgc_counter__', begin=0)

        # 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,
            name='__g_rampup_begin_step__',
            value=self._rampup_begin_step * 1.0,
            force_cpu=True)

        for param_var, grad_var in param_and_grads:
G
gongweibao 已提交
766
            var_numel = abs(reduce(lambda x, y: x * y, param_var.shape))
767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845
            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 :
                continue

            u_var = tensor.create_global_var(
                shape=param_var.shape,
                dtype=param_var.dtype,
                persistable=True,
                name=param_var.name + "__dgc_u__",
                value=0.0)
            v_var = tensor.create_global_var(
                shape=param_var.shape,
                dtype=param_var.dtype,
                persistable=True,
                name=param_var.name + "__dgc_v__",
                value=0.0)

            k_var = tensor.create_global_var(
                shape=[1],
                dtype=param_var.dtype,
                persistable=True,
                name=param_var.name + "__dgc_k__",
                value=0.0,
                force_cpu=True)

            encoded_var = tensor.create_global_var(
                shape=[1],
                dtype=param_var.dtype,
                persistable=True,
                name=param_var.name + "__dgc_encoded__",
                value=0.0,
                force_cpu=False)

            # 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,
                         encoded_var)

    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:
            name = unique_name.generate(".".join([helper.name, 'tmp']))

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

        helper.append_op(
G
gongweibao 已提交
846
            type="dgc_clip_by_norm",
847 848 849 850 851 852 853 854 855 856 857 858
            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
859
                x=grad_var, max_norm=clip_norm, name=grad_var.name)
860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894

    def _dgc_op(self, param_var, clip_var, grad_var, u_var, v_var, k_var,
                encoded_var):
        block = framework.default_main_program().global_block()
        op_maker = core.op_proto_and_checker_maker
        dgc_op = block.append_op(
            type="dgc",
            inputs={
                "U": u_var,
                "V": v_var,
                "Grad": clip_var,
                "current_step": self._global_step_var
            },
            outputs={
                "U_out": u_var,
                "V_out": v_var,
                "EncodeGrad": encoded_var,
                "k": k_var,
                "Grad_out": grad_var
            },
            attrs={
                "m": self._momentum,
                "sparsity": self._sparsity,
                "use_nesterov": self._use_nesterov,
                "rampup_begin_step": float(self._rampup_begin_step),
                "rampup_step": float(self._rampup_step)
            },
            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])


895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918
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

    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.
        momentum (float): momentum factor
        lars_coeff (float): defines how much we trust the layer to change its weights.
        lars_weight_decay (float): weight decay coefficient for decaying using LARS.
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
M
minqiyang 已提交
919

920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974

    Examples:
        .. code-block:: python

            optimizer = fluid.optimizer.LarsMomentum(learning_rate=0.2, momentum=0.1, lars_weight_decay=0.001)
            optimizer.minimize(cost)
    """
    _velocity_acc_str = "velocity"

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

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

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

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

        velocity_acc = self._get_accumulator(self._velocity_acc_str,
                                             param_and_grad[0])
        # create the momentum optimize op
        momentum_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "Velocity": velocity_acc,
                "LearningRate": self._create_param_lr(param_and_grad)
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "VelocityOut": velocity_acc
            },
            attrs={
                "mu": self._momentum,
                "lars_coeff": self._lars_coeff,
                "lars_weight_decay": self._lars_weight_decay
M
minqiyang 已提交
975 976
            },
            stop_gradient=True)
977 978 979 980

        return momentum_op


981
class AdagradOptimizer(Optimizer):
Q
qiaolongfei 已提交
982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001
    """
    **Adaptive Gradient Algorithm (Adagrad)**

    The update is done as follows:

    .. math::

        moment\_out &= moment + grad * grad

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

    The original paper(http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf)
    does not have the epsilon attribute. It is added here in our implementation
    as also proposed here: http://cs231n.github.io/neural-networks-3/#ada
    for numerical stability to avoid the division by zero error.

    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.
        epsilon (float): a small float value for numerical stability.
X
Xin Pan 已提交
1002 1003 1004
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
X
xuezhong 已提交
1005
        initial_accumulator_value (float): Initial value for moment accumulator.
Q
qiaolongfei 已提交
1006 1007 1008 1009 1010 1011

    Examples:
        .. code-block:: python

            optimizer = fluid.optimizer.Adagrad(learning_rate=0.2)
            optimizer.minimize(cost)
1012 1013 1014
    """
    _moment_acc_str = "moment"

X
Xin Pan 已提交
1015 1016 1017 1018
    def __init__(self,
                 learning_rate,
                 epsilon=1.0e-6,
                 regularization=None,
1019
                 name=None,
X
xuezhong 已提交
1020
                 initial_accumulator_value=0.0):
1021 1022
        assert learning_rate is not None
        assert epsilon is not None
Q
Qiao Longfei 已提交
1023
        super(AdagradOptimizer, self).__init__(
X
Xin Pan 已提交
1024 1025 1026
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
1027 1028
        self.type = "adagrad"
        self._epsilon = epsilon
1029
        self.initial_accumulator_value = initial_accumulator_value
1030 1031 1032 1033 1034

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

        for p in parameters:
Q
Qiao Longfei 已提交
1035
            self._add_accumulator(self._moment_acc_str, p)
1036 1037 1038 1039 1040 1041

    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])
1042 1043 1044 1045 1046 1047 1048 1049 1050 1051
        startup_block = framework.default_startup_program().global_block()
        startup_block.append_op(
            type='fill_constant',
            inputs={},
            outputs={'Out': [moment_acc]},
            attrs={
                'dtype': moment_acc.dtype,
                'value': self.initial_accumulator_value,
                'shape': moment_acc.shape,
            })
1052

1053
        # Create the adagrad optimizer op
1054 1055 1056 1057 1058 1059
        adagrad_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "Moment": moment_acc,
1060
                "LearningRate": self._create_param_lr(param_and_grad)
1061 1062 1063
            },
            outputs={"ParamOut": param_and_grad[0],
                     "MomentOut": moment_acc},
M
minqiyang 已提交
1064 1065
            attrs={"epsilon": self._epsilon},
            stop_gradient=True)
1066 1067

        return adagrad_op
1068 1069 1070


class AdamOptimizer(Optimizer):
Q
qiaolongfei 已提交
1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097
    """
    This implements the Adam optimizer from Section 2 of the Adam
    paper : https://arxiv.org/abs/1412.6980.
    Adam is a first-order gradient-based optimization method based on
    adaptive estimates of lower-order moments.

    Adam updates:

    .. 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}

    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.
        beta1 (float): The exponential decay rate for the 1st moment estimates.
        beta2 (float): The exponential decay rate for the 2nd moment estimates.
        epsilon (float): a small float value for numerical stability.
1098
        regularization: A Regularizer, such as fluid.regularizer.L2DecayRegularizer.
X
Xin Pan 已提交
1099
        name: A optional name prefix.
1100 1101 1102 1103 1104 1105
        lazy_mode(bool: false): 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 is the 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.
Q
qiaolongfei 已提交
1106 1107 1108 1109 1110 1111 1112

    Examples:
        .. code-block:: python

            optimizer = fluid.optimizer.Adam(learning_rate=0.2)
            optimizer.minimize(cost)

1113 1114 1115
    """
    _moment1_acc_str = "moment1"
    _moment2_acc_str = "moment2"
Q
qiaolongfei 已提交
1116 1117
    _beta1_pow_acc_str = "beta1_pow_acc"
    _beta2_pow_acc_str = "beta2_pow_acc"
1118 1119 1120 1121 1122

    def __init__(self,
                 learning_rate=0.001,
                 beta1=0.9,
                 beta2=0.999,
1123
                 epsilon=1e-8,
X
Xin Pan 已提交
1124
                 regularization=None,
Q
Qiao Longfei 已提交
1125
                 name=None,
Q
Qiao Longfei 已提交
1126
                 lazy_mode=False):
1127 1128 1129 1130
        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
1131
        super(AdamOptimizer, self).__init__(
X
Xin Pan 已提交
1132 1133 1134
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
1135 1136 1137 1138
        self.type = "adam"
        self._beta1 = beta1
        self._beta2 = beta2
        self._epsilon = epsilon
Q
Qiao Longfei 已提交
1139
        self._lazy_mode = lazy_mode
1140 1141 1142 1143 1144 1145

    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
1146 1147
            self._add_accumulator(self._moment1_acc_str, p)
            self._add_accumulator(self._moment2_acc_str, p)
Q
qiaolongfei 已提交
1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159
            self._add_accumulator(
                name=self._beta1_pow_acc_str,
                param=p,
                dtype='float32',
                fill_value=self._beta1,
                shape=[1])
            self._add_accumulator(
                name=self._beta2_pow_acc_str,
                param=p,
                dtype='float32',
                fill_value=self._beta2,
                shape=[1])
1160 1161 1162 1163 1164 1165 1166 1167

    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
1168 1169 1170 1171 1172
        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])

1173
        # create the adam optimize op
1174 1175 1176 1177 1178
        adam_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
1179
                "LearningRate": self._create_param_lr(param_and_grad),
1180 1181
                "Moment1": moment1,
                "Moment2": moment2,
Q
qiaolongfei 已提交
1182 1183
                "Beta1Pow": beta1_pow_acc,
                "Beta2Pow": beta2_pow_acc
1184 1185 1186 1187 1188 1189 1190 1191 1192
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "Moment1Out": moment1,
                "Moment2Out": moment2
            },
            attrs={
                "beta1": self._beta1,
                "beta2": self._beta2,
Q
Qiao Longfei 已提交
1193
                "epsilon": self._epsilon,
1194 1195
                "lazy_mode": self._lazy_mode,
                "min_row_size_to_use_multithread": 1000
M
minqiyang 已提交
1196 1197
            },
            stop_gradient=True)
1198 1199 1200

        return adam_op

1201
    def _finish_update(self, block, param_and_grads):
1202 1203 1204
        """Update Beta1 and Beta2 Power accumulators
        """
        assert isinstance(block, framework.Block)
Q
Qiao Longfei 已提交
1205
        main_block = block.program.global_block()
1206 1207 1208
        for param, grad in param_and_grads:
            if grad is None:
                continue
X
Xin Pan 已提交
1209 1210
            with param.block.program._optimized_guard(
                [param, grad]), name_scope("optimizer"):
1211 1212 1213 1214 1215 1216 1217 1218
                beta1_pow_acc = self._get_accumulator(self._beta1_pow_acc_str,
                                                      param)
                beta2_pow_acc = self._get_accumulator(self._beta2_pow_acc_str,
                                                      param)
                main_block.append_op(
                    type="scale",
                    inputs={"X": beta1_pow_acc},
                    outputs={"Out": beta1_pow_acc},
M
minqiyang 已提交
1219 1220
                    attrs={"scale": self._beta1},
                    stop_gradient=True)
1221 1222 1223 1224 1225

                main_block.append_op(
                    type="scale",
                    inputs={"X": beta2_pow_acc},
                    outputs={"Out": beta2_pow_acc},
M
minqiyang 已提交
1226 1227
                    attrs={"scale": self._beta2},
                    stop_gradient=True)
1228 1229 1230


class AdamaxOptimizer(Optimizer):
Q
qiaolongfei 已提交
1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260
    """
    We implement the Adamax optimizer from Section 7 of the Adam
    paper: https://arxiv.org/abs/1412.6980. Adamax is a variant of the
    Adam algorithm based on the infinity norm.

    Adamax updates:

    .. 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}


    The original paper does not have an epsilon attribute.
    However, it is added here for numerical stability to prevent the
    division by 0 error.

    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.
        beta1 (float): The exponential decay rate for the 1st moment estimates.
        beta2 (float): The exponential decay rate for the 2nd moment estimates.
        epsilon (float): a small float value for numerical stability.
X
Xin Pan 已提交
1261 1262 1263
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
Q
qiaolongfei 已提交
1264 1265 1266 1267 1268 1269

    Examples:
        .. code-block:: python

            optimizer = fluid.optimizer.Adamax(learning_rate=0.2)
            optimizer.minimize(cost)
C
chengduo 已提交
1270 1271 1272

    Notes:
       Currently, AdamaxOptimizer doesn't support sparse parameter optimization.
1273 1274 1275
    """
    _moment_acc_str = "moment"
    _inf_norm_acc_str = "inf_norm"
Q
qiaolongfei 已提交
1276
    _beta1_pow_acc_str = "beta1_pow_acc"
1277 1278 1279 1280 1281

    def __init__(self,
                 learning_rate=0.001,
                 beta1=0.9,
                 beta2=0.999,
1282
                 epsilon=1e-8,
X
Xin Pan 已提交
1283 1284
                 regularization=None,
                 name=None):
1285 1286 1287 1288
        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
1289
        super(AdamaxOptimizer, self).__init__(
X
Xin Pan 已提交
1290 1291 1292
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
1293 1294 1295 1296 1297 1298 1299 1300
        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
1301 1302
            self._add_accumulator(self._moment_acc_str, p)
            self._add_accumulator(self._inf_norm_acc_str, p)
Q
qiaolongfei 已提交
1303 1304 1305 1306 1307 1308
            self._add_accumulator(
                name=self._beta1_pow_acc_str,
                param=p,
                dtype='float32',
                fill_value=self._beta1,
                shape=[1])
1309 1310 1311 1312 1313 1314 1315

    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
1316 1317
        beta1_pow_acc = self._get_accumulator(self._beta1_pow_acc_str,
                                              param_and_grad[0])
1318 1319 1320 1321 1322 1323
        # create the adamax optimize op
        adamax_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
1324
                "LearningRate": self._create_param_lr(param_and_grad),
1325 1326
                "Moment": moment,
                "InfNorm": inf_norm,
Q
qiaolongfei 已提交
1327
                "Beta1Pow": beta1_pow_acc
1328 1329 1330 1331 1332 1333 1334 1335 1336 1337
            },
            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
1338 1339
            },
            stop_gradient=True)
1340 1341 1342

        return adamax_op

1343
    def _finish_update(self, block, parameters_and_grads):
1344 1345 1346
        """Update Beta1 Power accumulator
        """
        assert isinstance(block, framework.Block)
Q
Qiao Longfei 已提交
1347
        main_block = block.program.global_block()
1348 1349 1350
        for param, grad in parameters_and_grads:
            if grad is None:
                continue
X
Xin Pan 已提交
1351 1352
            with param.block.program._optimized_guard(
                [param, grad]), name_scope('adamx'):
1353 1354 1355 1356 1357 1358
                beta1_pow_acc = self._get_accumulator(self._beta1_pow_acc_str,
                                                      param)
                main_block.append_op(
                    type="scale",
                    inputs={"X": beta1_pow_acc},
                    outputs={"Out": beta1_pow_acc},
M
minqiyang 已提交
1359 1360
                    attrs={"scale": self._beta1},
                    stop_gradient=True)
1361 1362 1363


class DecayedAdagradOptimizer(Optimizer):
1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385
    """
    **Decayed Adagrad Optimizer**

    The original paper(http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf)

    The update is done as follows:

    .. math::

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

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

    The original paper(http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf)
    does not have an epsilon attribute. It is added here for numerical
    stability to avoid the division by zero error.

    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.
        decay (float): decay rate.
        epsilon (float): a small float value for numerical stability.
X
Xin Pan 已提交
1386 1387 1388
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
1389 1390 1391 1392 1393 1394

    Examples:
        .. code-block:: python

            optimizer = fluid.optimizer.DecayedAdagrad(learning_rate=0.2)
            optimizer.minimize(cost)
C
chengduo 已提交
1395 1396 1397

    Notes:
       Currently, DecayedAdagradOptimizer doesn't support sparse parameter optimization.
1398 1399 1400
    """
    _moment_acc_str = "moment"

X
Xin Pan 已提交
1401 1402 1403 1404 1405 1406
    def __init__(self,
                 learning_rate,
                 decay=0.95,
                 epsilon=1.0e-6,
                 regularization=None,
                 name=None):
1407 1408 1409 1410
        assert learning_rate is not None
        assert decay is not None
        assert epsilon is not None

Q
Qiao Longfei 已提交
1411
        super(DecayedAdagradOptimizer, self).__init__(
X
Xin Pan 已提交
1412 1413 1414
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441
        self.type = "decayed_adagrad"
        self._decay = decay
        self._epsilon = epsilon

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

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

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

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

        # Create the decayed adagrad optimizer op
        decayed_adagrad_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "Moment": moment_acc,
                "LearningRate": self._create_param_lr(param_and_grad)
            },
            outputs={"ParamOut": param_and_grad[0],
                     "MomentOut": moment_acc},
M
minqiyang 已提交
1442 1443
            attrs={"epsilon": self._epsilon},
            stop_gradient=True)
1444 1445

        return decayed_adagrad_op
1446 1447


1448
class AdadeltaOptimizer(Optimizer):
1449 1450
    """
    **Adadelta Optimizer**
Q
qiaolongfei 已提交
1451

1452
    Simple Adadelta optimizer with average squared grad state and
1453
    average squared update state.
1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465
    The details of adadelta please refer to this
    `ADADELTA: AN ADAPTIVE LEARNING RATE METHOD
    <http://www.matthewzeiler.com/pubs/googleTR2012/googleTR2012.pdf>`_.

    ..  math::

        E(g_t^2) &= \\rho * E(g_{t-1}^2) + (1-\\rho) * g^2 \\\\
        learning\\_rate &= sqrt( ( E(dx_{t-1}^2) + \\epsilon ) / ( \\
                          E(g_t^2) + \\epsilon ) ) \\\\
        E(dx_t^2) &= \\rho * E(dx_{t-1}^2) + (1-\\rho) * (-g*learning\\_rate)^2

    Args:
Q
qiaolongfei 已提交
1466
        learning_rate(float): global learning rate
1467 1468
        rho(float): rho in equation
        epsilon(float): epsilon in equation
X
Xin Pan 已提交
1469 1470 1471
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
1472 1473 1474 1475 1476 1477 1478

    Examples:
        .. code-block:: python

            optimizer = fluid.optimizer.Adadelta(
                learning_rate=0.0003, epsilon=1.0e-6, rho=0.95)
            _, params_grads = optimizer.minimize(cost)
C
chengduo 已提交
1479 1480 1481

    Notes:
       Currently, AdadeltaOptimizer doesn't support sparse parameter optimization.
1482
    """
1483

1484 1485 1486
    _avg_squared_grad_acc_str = "_avg_squared_grad"
    _avg_squared_update_acc_str = "_avg_squared_update"

X
Xin Pan 已提交
1487 1488 1489 1490 1491 1492
    def __init__(self,
                 learning_rate,
                 epsilon=1.0e-6,
                 rho=0.95,
                 regularization=None,
                 name=None):
1493 1494 1495 1496 1497 1498
        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.")
1499
        super(AdadeltaOptimizer, self).__init__(
X
Xin Pan 已提交
1500 1501 1502
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
1503 1504 1505 1506 1507
        self.type = "adadelta"
        self._epsilon = epsilon
        self._rho = rho

    def _create_accumulators(self, block, parameters):
1508 1509
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")
1510 1511 1512 1513 1514 1515

        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):
1516 1517
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")
1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538

        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
1539 1540
                   "rho": self._rho},
            stop_gradient=True)
1541 1542 1543 1544

        return adadelta_op


Q
qingqing01 已提交
1545 1546 1547 1548 1549 1550 1551 1552 1553 1554
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
1555
        r(w, t) & = \\rho r(w, t-1) + (1 - \\rho)(\\nabla Q_{i}(w))^2
Q
qingqing01 已提交
1556 1557 1558 1559

        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
1560
    each weight. Then dividing the gradient by :math:`sqrt{v(w,t)}`.
Q
qingqing01 已提交
1561 1562 1563 1564 1565 1566

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

    ..  math::

Q
qiaolongfei 已提交
1567
        r(w, t) & = \\rho r(w, t-1) + (1 - \\rho)(\\nabla Q_{i}(w))^2
Q
qingqing01 已提交
1568

1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582
        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
1583 1584 1585 1586
            \\epsilon}} \\nabla Q_{i}(w)

        w & = w - v(w, t)

Q
qiaolongfei 已提交
1587
    where, :math:`\\rho` is a hyperparameter and typical values are 0.9, 0.95
Q
qingqing01 已提交
1588 1589 1590 1591 1592 1593
    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.


    Args:
Q
qiaolongfei 已提交
1594
        learning_rate(float): global learning rate.
Q
qingqing01 已提交
1595 1596 1597
        rho(float): rho is :math: `\\rho` in equation, set 0.95 by default.
        epsilon(float): :math: `\\epsilon` in equation is smoothing term to
            avoid division by zero, set 1e-6 by default.
Q
qiaolongfei 已提交
1598
        momentum(float): :math:`\\beta` in equation is the momentum term,
Q
qingqing01 已提交
1599
            set 0.0 by default.
1600 1601 1602 1603
        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.
X
Xin Pan 已提交
1604 1605 1606
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
Q
qingqing01 已提交
1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619

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

    Examples:
          .. code-block:: python

              optimizer = fluid.optimizer.RMSProp(0.0001)
              _, params_grads = optimizer.minimize(cost)
    """

    _momentum_acc_str = "momentum"
    _mean_square_acc_str = "mean_square"
1620
    _mean_grad_acc_str = "mean_grad"
Q
qingqing01 已提交
1621 1622 1623 1624 1625 1626

    def __init__(self,
                 learning_rate,
                 rho=0.95,
                 epsilon=1.0e-6,
                 momentum=0.0,
1627
                 centered=False,
X
Xin Pan 已提交
1628 1629
                 regularization=None,
                 name=None):
Q
qingqing01 已提交
1630
        super(RMSPropOptimizer, self).__init__(
X
Xin Pan 已提交
1631 1632 1633
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
Q
qingqing01 已提交
1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646
        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
1647
        self._centered = centered
Q
qingqing01 已提交
1648 1649 1650 1651 1652 1653 1654 1655

    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)
1656
            self._add_accumulator(self._mean_grad_acc_str, p)
Q
qingqing01 已提交
1657 1658 1659 1660 1661 1662 1663 1664 1665

    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])
1666 1667
        mean_grad_acc = self._get_accumulator(self._mean_grad_acc_str,
                                              param_and_grad[0])
Q
qingqing01 已提交
1668 1669 1670 1671 1672 1673 1674
        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,
1675
                "MeanGrad": mean_grad_acc,
Q
qingqing01 已提交
1676 1677 1678 1679 1680
                "LearningRate": self._create_param_lr(param_and_grad),
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "MomentOut": momentum_acc,
1681 1682
                "MeanSquareOut": mean_square_acc,
                "MeanGradOut": mean_grad_acc
Q
qingqing01 已提交
1683 1684 1685 1686
            },
            attrs={
                "epsilon": self._epsilon,
                "decay": self._rho,
1687 1688
                "momentum": self._momentum,
                "centered": self._centered
M
minqiyang 已提交
1689 1690
            },
            stop_gradient=True)
Q
qingqing01 已提交
1691 1692 1693 1694

        return rmsprop_op


Q
qiaolongfei 已提交
1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 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 1734 1735 1736
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

    Args:
        learning_rate (float|Variable): global learning rate.
M
minqiyang 已提交
1737 1738 1739
        l1 (float): L1 regularization strength.
        l2 (float): L2 regularization strength.
        lr_power (float): Learning Rate Power.
X
Xin Pan 已提交
1740 1741 1742
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
Q
qiaolongfei 已提交
1743 1744 1745 1746 1747 1748 1749 1750 1751

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

    Examples:
          .. code-block:: python

              optimizer = fluid.optimizer.Ftrl(0.0001)
              _, params_grads = optimizer.minimize(cost)
C
chengduo 已提交
1752 1753 1754

    Notes:
       Currently, FtrlOptimizer doesn't support sparse parameter optimization.
Q
qiaolongfei 已提交
1755 1756 1757 1758 1759
    """

    _squared_acc_str = "squared"
    _linear_acc_str = "linear"

X
Xin Pan 已提交
1760 1761 1762 1763 1764 1765 1766
    def __init__(self,
                 learning_rate,
                 l1=0.0,
                 l2=0.0,
                 lr_power=-0.5,
                 regularization=None,
                 name=None):
Q
qiaolongfei 已提交
1767
        super(FtrlOptimizer, self).__init__(
X
Xin Pan 已提交
1768 1769 1770
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
Q
qiaolongfei 已提交
1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810
        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
1811 1812
                   "lr_power": self._lr_power},
            stop_gradient=True)
Q
qiaolongfei 已提交
1813 1814 1815 1816

        return ftrl_op


1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830
# 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
DecayedAdagrad = DecayedAdagradOptimizer
1831
Adadelta = AdadeltaOptimizer
Q
qingqing01 已提交
1832
RMSProp = RMSPropOptimizer
Q
qiaolongfei 已提交
1833
Ftrl = FtrlOptimizer
1834
LarsMomentum = LarsMomentumOptimizer
1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849


class ModelAverage(Optimizer):
    """Accumulate the average of parameters whtin sliding window. The average
    result will be saved in temporary variables which can be applied to
    parameter variables of current model by calling 'apply()' method. And the
    'restore()' method is used to restored the parameter values of current model.

    The size of average window is determined by average_window_rate,
    min_average_window, max_average_window and current update times.

    Args:
        average_window_rate: The rate of average window.
        min_average_window: The minimum size of average window.
        max_average_window: The maximum size of average window.
X
Xin Pan 已提交
1850 1851 1852
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
1853
    Examples:
Q
qiaolongfei 已提交
1854 1855 1856

      .. code-block:: python

1857
        optimizer = fluid.optimizer.Momentum()
1858 1859
        optimizer.minimize(cost)
        model_average = fluid.optimizer.ModelAverage(0.15,
1860 1861 1862 1863 1864
                                                min_average_window=10000,
                                                max_average_window=20000)
        for pass_id in range(args.pass_num):
            for data in train_reader():
                exe.run(fluid.default_main_program()...)
1865 1866 1867 1868

            with model_average.apply(exe):
                for data in test_reader():
                    exe.run(inference_program...)
1869 1870 1871
    """

    def __init__(self,
W
wanghaoshuang 已提交
1872
                 average_window_rate,
1873 1874
                 min_average_window=10000,
                 max_average_window=10000,
X
Xin Pan 已提交
1875 1876 1877 1878
                 regularization=None,
                 name=None):
        super(ModelAverage, self).__init__(
            0.0, regularization=regularization, name=name)
1879 1880 1881
        self.average_window = average_window_rate
        self.min_average_window = min_average_window
        self.max_average_window = max_average_window
1882

1883
        self.params_grads = []
1884 1885
        for param in framework.default_main_program().global_block(
        ).all_parameters():
1886
            if param.do_model_average != False:
1887 1888 1889 1890
                grad = param.block.create_var(
                    name=unique_name.generate(".".join([param.name, 'tmp'])),
                    dtype=param.dtype,
                    persistable=False,
W
wanghaoshuang 已提交
1891
                    stop_gradient=True)
1892
                self.params_grads.append((param, grad))
1893

1894
        for param, grad in self.params_grads:
1895 1896
            if grad is None:
                continue
X
Xin Pan 已提交
1897 1898
            with param.block.program._optimized_guard(
                [param, grad]), name_scope('move_average'):
1899
                self._append_average_accumulate_op(param)
1900

1901 1902 1903 1904
        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:
1905
                self._add_average_apply_op(block, param_grad)
1906 1907 1908 1909 1910

        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:
1911
                self._add_average_restore_op(block, param_grad)
1912

1913
    def _add_average_apply_op(self, block, param_grad):
L
Luo Tao 已提交
1914 1915 1916 1917 1918 1919
        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(
1920
            self._get_accumulator('num_accumulates', param))
L
Luo Tao 已提交
1921
        old_num_accumulates = block._clone_variable(
1922
            self._get_accumulator('old_num_accumulates', param))
L
Luo Tao 已提交
1923
        num_updates = block._clone_variable(
1924 1925 1926 1927 1928 1929
            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
1930 1931 1932 1933
        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
1934
        ops._elementwise_div(x=sum, y=tmp, out=param)
1935 1936

    def _add_average_restore_op(self, block, param_grad):
L
Luo Tao 已提交
1937 1938
        param = block._clone_variable(param_grad[0])
        grad = block._clone_variable(param_grad[1])
1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975
        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
1976 1977
            },
            stop_gradient=True)
1978

S
rename  
sneaxiy 已提交
1979
    @signature_safe_contextmanager
1980
    def apply(self, executor, need_restore=True):
1981 1982
        """Apply average values to parameters of current model.
        """
1983 1984 1985 1986 1987 1988
        executor.run(self.apply_program)
        try:
            yield
        finally:
            if need_restore:
                self.restore(executor)
1989 1990 1991 1992

    def restore(self, executor):
        """Restore parameter values of current model.
        """
1993
        executor.run(self.restore_program)