optimizer.py 49.3 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
W
wanghaoshuang 已提交
16
import re
17
from collections import defaultdict
18
from paddle.fluid.framework import Program, Variable, name_scope
19 20 21 22 23 24 25 26 27
from . import framework
from . import layers
from .backward import append_backward
from .framework import program_guard
from . import unique_name
from .initializer import Constant
from .layer_helper import LayerHelper
from .regularizer import append_regularization_ops
from .clip import append_gradient_clip_ops, error_clip_callback
28
from contextlib import contextmanager
S
sneaxiy 已提交
29
from .layers import ops
30

31
__all__ = [
Q
qiaolongfei 已提交
32
    'SGD', 'Momentum', 'Adagrad', 'Adam', 'Adamax', 'DecayedAdagrad', 'Ftrl',
33
    'SGDOptimizer', 'MomentumOptimizer', 'AdagradOptimizer', 'AdamOptimizer',
W
weixing02 已提交
34
    'AdamaxOptimizer', 'DecayedAdagradOptimizer', 'RMSPropOptimizer',
Y
yuyang18 已提交
35
    'FtrlOptimizer', 'Adadelta', 'ModelAverage', 'RMSPropOptimizer'
36
]
Q
Qiao Longfei 已提交
37 38 39 40 41 42


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

    Define the common interface of an optimizer.
43 44
    User should not use this class directly,
    but need to use one of it's implementation.
Q
Qiao Longfei 已提交
45 46
    """

X
Xin Pan 已提交
47
    def __init__(self, learning_rate, regularization=None, name=None):
48 49
        if not isinstance(learning_rate, float) and \
                not isinstance(learning_rate, framework.Variable):
Q
qiaolongfei 已提交
50
            raise TypeError("learning rate should be float or Variable")
W
whs 已提交
51
        self._name = name
D
dzhwinter 已提交
52
        self.regularization = regularization
53
        self._learning_rate = learning_rate
D
dzhwinter 已提交
54 55
        # the learning rate type should be inferenced from loss
        self._dtype = None
56 57
        # each program should have a independent learning rate
        # program -> Variable(learning_rate)
Q
qiaolongfei 已提交
58
        self._learning_rate_map = dict()
59 60 61
        if isinstance(self._learning_rate, framework.Variable):
            self._learning_rate_map[framework.default_main_program(
            )] = self._learning_rate
62 63 64 65 66
        # 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
67
        self.helper = None
Q
Qiao Longfei 已提交
68

Q
Qiao Longfei 已提交
69
    def _create_global_learning_rate(self):
Y
yuyang18 已提交
70
        lr = self._global_learning_rate()
Q
Qiao Longfei 已提交
71

72 73 74 75
        if isinstance(lr, framework.Variable):
            return
        else:
            if not isinstance(self._learning_rate, float):
Q
qiaolongfei 已提交
76
                raise TypeError(
77 78
                    "learning rate variable is create outside optimizer,"
                    "can not create new learning rate variable for new program")
Q
Qiao Longfei 已提交
79

80 81 82 83 84 85
        # 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),
D
dzhwinter 已提交
86
            dtype='float32' if self._dtype == None else self._dtype,
87 88
            persistable=True)

Y
yuyang18 已提交
89
    def _global_learning_rate(self, program=None):
Q
Qiao Longfei 已提交
90 91 92 93
        """
        get global decayed learning rate
        :return:
        """
94 95
        if program is None:
            program = framework.default_main_program()
Q
qiaolongfei 已提交
96
        return self._learning_rate_map.get(program, None)
Q
Qiao Longfei 已提交
97

Q
Qiao Longfei 已提交
98 99 100 101 102
    def _append_optimize_op(self, block, param_and_grad):
        """ append optimize operator to block and return all the added optimize_op
        """
        raise NotImplementedError()

103 104 105 106
    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
107
        if type(param_lr) == Variable:
108
            print("returns updated param lr ", param_lr)
W
Wu Yi 已提交
109
            return param_lr
Q
qiaolongfei 已提交
110
        else:
W
Wu Yi 已提交
111
            if param_lr == 1.0:
Y
yuyang18 已提交
112
                return self._global_learning_rate()
W
Wu Yi 已提交
113
            else:
Y
yuyang18 已提交
114
                return self._global_learning_rate() * param_lr
115 116 117 118 119 120 121

    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
122
        """
123 124
        pass

125
    def _finish_update(self, block, parameters_and_grads):
126 127 128 129 130 131 132 133
        """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
134
            None
135 136 137
        """
        pass

138 139 140 141 142 143
    def _add_accumulator(self,
                         name,
                         param,
                         dtype=None,
                         fill_value=0.0,
                         shape=None):
144 145 146 147 148 149 150 151 152
        """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
153 154
        if self._name is not None:
            name = self._name + "_" + name
155 156
        if (name in self._accumulators and
                param.name in self._accumulators[name]):
157
            raise Exception("Accumulator {} already exists for parameter {}".
158
                            format(name, param.name))
159 160
        if shape == None:
            shape = param.shape
Q
Qiao Longfei 已提交
161 162
        assert isinstance(self.helper, LayerHelper)
        var = self.helper.create_global_variable(
Y
Yu Yang 已提交
163
            name=unique_name.generate(name),
Q
Qiao Longfei 已提交
164
            persistable=True,
F
fengjiayi 已提交
165
            dtype=dtype or param.dtype,
Q
Qiao Longfei 已提交
166
            type=param.type,
167
            shape=shape)
Q
Qiao Longfei 已提交
168
        self.helper.set_variable_initializer(
169
            var, initializer=Constant(value=float(fill_value)))
Q
Qiao Longfei 已提交
170
        self._accumulators[name][param.name] = var
171
        return var
172 173 174 175 176 177 178 179 180 181 182

    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
183 184
        if self._name is not None:
            name = self._name + "_" + name
185 186 187 188 189 190
        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]

Y
yuyang18 已提交
191 192 193 194
    def _create_optimization_pass(self,
                                  parameters_and_grads,
                                  loss,
                                  startup_program=None):
Q
Qiao Longfei 已提交
195 196 197
        """Add optimization operators to update gradients to variables.

        Args:
Q
qiaolongfei 已提交
198 199 200
          loss(Variable): the target that this optimization is for.
          parameters_and_grads(list(tuple(Variable, Variable))):
          a list of (variable, gradient) pair to update.
Q
Qiao Longfei 已提交
201 202

        Returns:
203 204 205 206
          return_op_list: a list of operators that will complete one step of
          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
207
        """
208 209 210 211 212
        # 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
213
        # for parameters and extend _finish_update method to add custom ops.
214 215

        # Create any accumulators
Q
Qiao Longfei 已提交
216
        program = loss.block.program
D
dzhwinter 已提交
217
        self._dtype = loss.dtype
218
        with program_guard(program, startup_program):
Y
Yancey1989 已提交
219 220
            global_block = framework.default_main_program().global_block()
            start = len(global_block.ops)
221 222 223
            self.helper = LayerHelper(self.__class__.__name__)
            self._create_accumulators(loss.block,
                                      [p[0] for p in parameters_and_grads])
Q
Qiao Longfei 已提交
224
            self._create_global_learning_rate()
225 226 227

            optimize_ops = []
            for param_and_grad in parameters_and_grads:
228 229
                if param_and_grad[1] is None:
                    continue
W
Wu Yi 已提交
230
                with param_and_grad[0].block.program._optimized_guard(
231
                        param_and_grad), name_scope("optimizer"):
232
                    if param_and_grad[0].trainable is True:
Y
yuyang18 已提交
233 234 235
                        optimize_op = self._append_optimize_op(loss.block,
                                                               param_and_grad)
                        optimize_ops.append(optimize_op)
236 237 238

            # Get custom finish ops for subclasses
            # FIXME: Need to fix this once we figure out how to handle dependencies
239
            self._finish_update(loss.block, parameters_and_grads)
240

Y
Yancey1989 已提交
241
            end = len(global_block.ops)
W
Wu Yi 已提交
242
            return global_block._slice_ops(start, end)
Q
Qiao Longfei 已提交
243

Q
Qiao Longfei 已提交
244 245
    def minimize(self,
                 loss,
246
                 startup_program=None,
Q
Qiao Longfei 已提交
247 248
                 parameter_list=None,
                 no_grad_set=None):
Q
Qiao Longfei 已提交
249 250
        """Add operations to minimize `loss` by updating `parameter_list`.

F
fengjiayi 已提交
251
        This method combines interface `append_backward()` and
Q
Qiao Longfei 已提交
252 253
        `create_optimization_pass()` into one.
        """
F
fengjiayi 已提交
254
        params_grads = append_backward(loss, parameter_list, no_grad_set,
Y
Yang Yang 已提交
255
                                       [error_clip_callback])
Y
Yu Yang 已提交
256

Y
Yu Yang 已提交
257 258
        params_grads = sorted(params_grads, key=lambda x: x[0].name)

Y
Yu Yang 已提交
259 260
        params_grads = append_gradient_clip_ops(params_grads)

F
fengjiayi 已提交
261
        # Add regularization if any
D
dzhwinter 已提交
262 263
        params_grads = append_regularization_ops(params_grads,
                                                 self.regularization)
Y
Yu Yang 已提交
264

Y
yuyang18 已提交
265 266
        optimize_ops = self._create_optimization_pass(params_grads, loss,
                                                      startup_program)
T
typhoonzero 已提交
267
        return optimize_ops, params_grads
Q
Qiao Longfei 已提交
268 269 270


class SGDOptimizer(Optimizer):
Q
qiaolongfei 已提交
271 272 273 274 275 276 277 278 279 280
    """
    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
281 282 283
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
Q
qiaolongfei 已提交
284 285 286 287

    Examples:
        .. code-block:: python

Q
qiaolongfei 已提交
288
            sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.2)
Q
qiaolongfei 已提交
289
            sgd_optimizer.minimize(cost)
Q
Qiao Longfei 已提交
290 291
    """

X
Xin Pan 已提交
292
    def __init__(self, learning_rate, regularization=None, name=None):
Q
Qiao Longfei 已提交
293
        assert learning_rate is not None
Q
Qiao Longfei 已提交
294
        super(SGDOptimizer, self).__init__(
X
Xin Pan 已提交
295 296 297
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
Q
Qiao Longfei 已提交
298 299
        self.type = "sgd"

300 301
    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)
302

Q
Qiao Longfei 已提交
303 304 305 306 307 308
        # create the optimize op
        sgd_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
309
                "LearningRate": self._create_param_lr(param_and_grad)
Q
Qiao Longfei 已提交
310
            },
311
            outputs={"ParamOut": param_and_grad[0]})
Q
Qiao Longfei 已提交
312 313

        return sgd_op
314 315 316


class MomentumOptimizer(Optimizer):
Q
qiaolongfei 已提交
317 318 319 320 321 322 323 324 325 326 327 328 329 330
    """

    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):

331
        &\quad   param = param - (gradient + mu * velocity) * learning\_rate
Q
qiaolongfei 已提交
332 333 334

        & else:

Q
qiaolongfei 已提交
335
        &\quad   param = param - learning\_rate * velocity
Q
qiaolongfei 已提交
336 337 338 339 340 341

    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
342 343 344
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
Q
qiaolongfei 已提交
345 346 347 348

    Examples:
        .. code-block:: python

Q
qiaolongfei 已提交
349
            optimizer = fluid.optimizer.Momentum(learning_rate=0.2, momentum=0.1)
Q
qiaolongfei 已提交
350
            optimizer.minimize(cost)
351 352 353
    """
    _velocity_acc_str = "velocity"

X
Xin Pan 已提交
354 355 356 357 358 359
    def __init__(self,
                 learning_rate,
                 momentum,
                 use_nesterov=False,
                 regularization=None,
                 name=None):
360 361
        assert learning_rate is not None
        assert momentum is not None
Q
Qiao Longfei 已提交
362
        super(MomentumOptimizer, self).__init__(
X
Xin Pan 已提交
363 364 365
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
366 367
        self.type = "momentum"
        self._momentum = momentum
368
        self._use_nesterov = bool(use_nesterov)
369 370 371 372 373

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

        for p in parameters:
Q
Qiao Longfei 已提交
374
            self._add_accumulator(self._velocity_acc_str, p)
375 376 377 378 379 380 381 382 383 384 385 386 387

    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,
388
                "LearningRate": self._create_param_lr(param_and_grad)
389 390 391 392 393
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "VelocityOut": velocity_acc
            },
394
            attrs={"mu": self._momentum,
395
                   "use_nesterov": self._use_nesterov})
396 397

        return momentum_op
398 399 400


class AdagradOptimizer(Optimizer):
Q
qiaolongfei 已提交
401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420
    """
    **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
421 422 423
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
Q
qiaolongfei 已提交
424 425 426 427 428 429

    Examples:
        .. code-block:: python

            optimizer = fluid.optimizer.Adagrad(learning_rate=0.2)
            optimizer.minimize(cost)
430 431 432
    """
    _moment_acc_str = "moment"

X
Xin Pan 已提交
433 434 435 436 437
    def __init__(self,
                 learning_rate,
                 epsilon=1.0e-6,
                 regularization=None,
                 name=None):
438 439
        assert learning_rate is not None
        assert epsilon is not None
Q
Qiao Longfei 已提交
440
        super(AdagradOptimizer, self).__init__(
X
Xin Pan 已提交
441 442 443
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
444 445 446 447 448 449 450
        self.type = "adagrad"
        self._epsilon = epsilon

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

        for p in parameters:
Q
Qiao Longfei 已提交
451
            self._add_accumulator(self._moment_acc_str, p)
452 453 454 455 456 457 458

    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])

459
        # Create the adagrad optimizer op
460 461 462 463 464 465
        adagrad_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "Moment": moment_acc,
466
                "LearningRate": self._create_param_lr(param_and_grad)
467 468 469 470 471 472
            },
            outputs={"ParamOut": param_and_grad[0],
                     "MomentOut": moment_acc},
            attrs={"epsilon": self._epsilon})

        return adagrad_op
473 474 475


class AdamOptimizer(Optimizer):
Q
qiaolongfei 已提交
476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502
    """
    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.
X
Xin Pan 已提交
503 504 505
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
Q
qiaolongfei 已提交
506 507 508 509 510 511 512

    Examples:
        .. code-block:: python

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

513 514 515
    """
    _moment1_acc_str = "moment1"
    _moment2_acc_str = "moment2"
Q
qiaolongfei 已提交
516 517
    _beta1_pow_acc_str = "beta1_pow_acc"
    _beta2_pow_acc_str = "beta2_pow_acc"
518 519 520 521 522

    def __init__(self,
                 learning_rate=0.001,
                 beta1=0.9,
                 beta2=0.999,
523
                 epsilon=1e-8,
X
Xin Pan 已提交
524 525
                 regularization=None,
                 name=None):
526 527 528 529
        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
530
        super(AdamOptimizer, self).__init__(
X
Xin Pan 已提交
531 532 533
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
534 535 536 537 538 539 540 541 542 543
        self.type = "adam"
        self._beta1 = beta1
        self._beta2 = beta2
        self._epsilon = epsilon

    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
544 545
            self._add_accumulator(self._moment1_acc_str, p)
            self._add_accumulator(self._moment2_acc_str, p)
Q
qiaolongfei 已提交
546 547 548 549 550 551 552 553 554 555 556 557
            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])
558 559 560 561 562 563 564 565

    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
566 567 568 569 570
        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])

571
        # create the adam optimize op
572 573 574 575 576
        adam_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
577
                "LearningRate": self._create_param_lr(param_and_grad),
578 579
                "Moment1": moment1,
                "Moment2": moment2,
Q
qiaolongfei 已提交
580 581
                "Beta1Pow": beta1_pow_acc,
                "Beta2Pow": beta2_pow_acc
582 583 584 585 586 587 588 589 590 591 592 593 594 595
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "Moment1Out": moment1,
                "Moment2Out": moment2
            },
            attrs={
                "beta1": self._beta1,
                "beta2": self._beta2,
                "epsilon": self._epsilon
            })

        return adam_op

596
    def _finish_update(self, block, param_and_grads):
597 598 599
        """Update Beta1 and Beta2 Power accumulators
        """
        assert isinstance(block, framework.Block)
Q
Qiao Longfei 已提交
600
        main_block = block.program.global_block()
601 602 603
        for param, grad in param_and_grads:
            if grad is None:
                continue
W
Wu Yi 已提交
604
            with param.block.program._optimized_guard([param, grad]):
605 606 607 608 609 610 611 612 613 614 615 616 617 618 619
                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},
                    attrs={"scale": self._beta1})

                main_block.append_op(
                    type="scale",
                    inputs={"X": beta2_pow_acc},
                    outputs={"Out": beta2_pow_acc},
                    attrs={"scale": self._beta2})
620 621 622


class AdamaxOptimizer(Optimizer):
Q
qiaolongfei 已提交
623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652
    """
    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
653 654 655
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
Q
qiaolongfei 已提交
656 657 658 659 660 661

    Examples:
        .. code-block:: python

            optimizer = fluid.optimizer.Adamax(learning_rate=0.2)
            optimizer.minimize(cost)
C
chengduo 已提交
662 663 664

    Notes:
       Currently, AdamaxOptimizer doesn't support sparse parameter optimization.
665 666 667
    """
    _moment_acc_str = "moment"
    _inf_norm_acc_str = "inf_norm"
Q
qiaolongfei 已提交
668
    _beta1_pow_acc_str = "beta1_pow_acc"
669 670 671 672 673

    def __init__(self,
                 learning_rate=0.001,
                 beta1=0.9,
                 beta2=0.999,
674
                 epsilon=1e-8,
X
Xin Pan 已提交
675 676
                 regularization=None,
                 name=None):
677 678 679 680
        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
681
        super(AdamaxOptimizer, self).__init__(
X
Xin Pan 已提交
682 683 684
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
685 686 687 688 689 690 691 692
        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
693 694
            self._add_accumulator(self._moment_acc_str, p)
            self._add_accumulator(self._inf_norm_acc_str, p)
Q
qiaolongfei 已提交
695 696 697 698 699 700
            self._add_accumulator(
                name=self._beta1_pow_acc_str,
                param=p,
                dtype='float32',
                fill_value=self._beta1,
                shape=[1])
701 702 703 704 705 706 707

    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
708 709
        beta1_pow_acc = self._get_accumulator(self._beta1_pow_acc_str,
                                              param_and_grad[0])
710 711 712 713 714 715
        # create the adamax optimize op
        adamax_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
716
                "LearningRate": self._create_param_lr(param_and_grad),
717 718
                "Moment": moment,
                "InfNorm": inf_norm,
Q
qiaolongfei 已提交
719
                "Beta1Pow": beta1_pow_acc
720 721 722 723 724 725 726 727 728 729 730 731 732 733
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "MomentOut": moment,
                "InfNormOut": inf_norm
            },
            attrs={
                "beta1": self._beta1,
                "beta2": self._beta2,
                "epsilon": self._epsilon
            })

        return adamax_op

734
    def _finish_update(self, block, parameters_and_grads):
735 736 737
        """Update Beta1 Power accumulator
        """
        assert isinstance(block, framework.Block)
Q
Qiao Longfei 已提交
738
        main_block = block.program.global_block()
739 740 741
        for param, grad in parameters_and_grads:
            if grad is None:
                continue
W
Wu Yi 已提交
742
            with param.block.program._optimized_guard([param, grad]):
743 744 745 746 747 748 749
                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},
                    attrs={"scale": self._beta1})
750 751 752


class DecayedAdagradOptimizer(Optimizer):
753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774
    """
    **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
775 776 777
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
778 779 780 781 782 783

    Examples:
        .. code-block:: python

            optimizer = fluid.optimizer.DecayedAdagrad(learning_rate=0.2)
            optimizer.minimize(cost)
C
chengduo 已提交
784 785 786

    Notes:
       Currently, DecayedAdagradOptimizer doesn't support sparse parameter optimization.
787 788 789
    """
    _moment_acc_str = "moment"

X
Xin Pan 已提交
790 791 792 793 794 795
    def __init__(self,
                 learning_rate,
                 decay=0.95,
                 epsilon=1.0e-6,
                 regularization=None,
                 name=None):
796 797 798 799
        assert learning_rate is not None
        assert decay is not None
        assert epsilon is not None

Q
Qiao Longfei 已提交
800
        super(DecayedAdagradOptimizer, self).__init__(
X
Xin Pan 已提交
801 802 803
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
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
        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},
            attrs={"epsilon": self._epsilon})

        return decayed_adagrad_op
834 835


836
class AdadeltaOptimizer(Optimizer):
837 838
    """
    **Adadelta Optimizer**
Q
qiaolongfei 已提交
839

840
    Simple Adadelta optimizer with average squared grad state and
841
    average squared update state.
842 843 844 845 846 847 848 849 850 851 852 853
    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
854
        learning_rate(float): global learning rate
855 856
        rho(float): rho in equation
        epsilon(float): epsilon in equation
X
Xin Pan 已提交
857 858 859
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
860 861 862 863 864 865 866

    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
867 868 869

    Notes:
       Currently, AdadeltaOptimizer doesn't support sparse parameter optimization.
870
    """
871

872 873 874
    _avg_squared_grad_acc_str = "_avg_squared_grad"
    _avg_squared_update_acc_str = "_avg_squared_update"

X
Xin Pan 已提交
875 876 877 878 879 880
    def __init__(self,
                 learning_rate,
                 epsilon=1.0e-6,
                 rho=0.95,
                 regularization=None,
                 name=None):
881 882 883 884 885 886
        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.")
887
        super(AdadeltaOptimizer, self).__init__(
X
Xin Pan 已提交
888 889 890
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
891 892 893 894 895
        self.type = "adadelta"
        self._epsilon = epsilon
        self._rho = rho

    def _create_accumulators(self, block, parameters):
896 897
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")
898 899 900 901 902 903

        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):
904 905
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")
906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931

        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,
                   "rho": self._rho})

        return adadelta_op


Q
qingqing01 已提交
932 933 934 935 936 937 938 939 940 941
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
942
        r(w, t) & = \\rho r(w, t-1) + (1 - \\rho)(\\nabla Q_{i}(w))^2
Q
qingqing01 已提交
943 944 945 946

        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
947
    each weight. Then dividing the gradient by :math:`sqrt{v(w,t)}`.
Q
qingqing01 已提交
948 949 950 951 952 953

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

    ..  math::

Q
qiaolongfei 已提交
954
        r(w, t) & = \\rho r(w, t-1) + (1 - \\rho)(\\nabla Q_{i}(w))^2
Q
qingqing01 已提交
955

956 957 958 959 960 961 962 963 964 965 966 967 968 969
        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
970 971 972 973
            \\epsilon}} \\nabla Q_{i}(w)

        w & = w - v(w, t)

Q
qiaolongfei 已提交
974
    where, :math:`\\rho` is a hyperparameter and typical values are 0.9, 0.95
Q
qingqing01 已提交
975 976 977 978 979 980
    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
981
        learning_rate(float): global learning rate.
Q
qingqing01 已提交
982 983 984
        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
985
        momentum(float): :math:`\\beta` in equation is the momentum term,
Q
qingqing01 已提交
986
            set 0.0 by default.
987 988 989 990
        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
991 992 993
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
Q
qingqing01 已提交
994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006

    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"
1007
    _mean_grad_acc_str = "mean_grad"
Q
qingqing01 已提交
1008 1009 1010 1011 1012 1013

    def __init__(self,
                 learning_rate,
                 rho=0.95,
                 epsilon=1.0e-6,
                 momentum=0.0,
1014
                 centered=False,
X
Xin Pan 已提交
1015 1016
                 regularization=None,
                 name=None):
Q
qingqing01 已提交
1017
        super(RMSPropOptimizer, self).__init__(
X
Xin Pan 已提交
1018 1019 1020
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
Q
qingqing01 已提交
1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033
        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
1034
        self._centered = centered
Q
qingqing01 已提交
1035 1036 1037 1038 1039 1040 1041 1042

    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)
1043
            self._add_accumulator(self._mean_grad_acc_str, p)
Q
qingqing01 已提交
1044 1045 1046 1047 1048 1049 1050 1051 1052

    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])
1053 1054
        mean_grad_acc = self._get_accumulator(self._mean_grad_acc_str,
                                              param_and_grad[0])
Q
qingqing01 已提交
1055 1056 1057 1058 1059 1060 1061
        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,
1062
                "MeanGrad": mean_grad_acc,
Q
qingqing01 已提交
1063 1064 1065 1066 1067
                "LearningRate": self._create_param_lr(param_and_grad),
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "MomentOut": momentum_acc,
1068 1069
                "MeanSquareOut": mean_square_acc,
                "MeanGradOut": mean_grad_acc
Q
qingqing01 已提交
1070 1071 1072 1073
            },
            attrs={
                "epsilon": self._epsilon,
                "decay": self._rho,
1074 1075
                "momentum": self._momentum,
                "centered": self._centered
Q
qingqing01 已提交
1076 1077 1078 1079 1080
            })

        return rmsprop_op


Q
qiaolongfei 已提交
1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125
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.
        l1 (float):
        l2 (float):
        lr_power (float):
X
Xin Pan 已提交
1126 1127 1128
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
Q
qiaolongfei 已提交
1129 1130 1131 1132 1133 1134 1135 1136 1137

    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
1138 1139 1140

    Notes:
       Currently, FtrlOptimizer doesn't support sparse parameter optimization.
Q
qiaolongfei 已提交
1141 1142 1143 1144 1145
    """

    _squared_acc_str = "squared"
    _linear_acc_str = "linear"

X
Xin Pan 已提交
1146 1147 1148 1149 1150 1151 1152
    def __init__(self,
                 learning_rate,
                 l1=0.0,
                 l2=0.0,
                 lr_power=-0.5,
                 regularization=None,
                 name=None):
Q
qiaolongfei 已提交
1153
        super(FtrlOptimizer, self).__init__(
X
Xin Pan 已提交
1154 1155 1156
            learning_rate=learning_rate,
            regularization=regularization,
            name=name)
Q
qiaolongfei 已提交
1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201
        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,
                   "lr_power": self._lr_power})

        return ftrl_op


1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215
# 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
1216
Adadelta = AdadeltaOptimizer
Q
qingqing01 已提交
1217
RMSProp = RMSPropOptimizer
Q
qiaolongfei 已提交
1218
Ftrl = FtrlOptimizer
1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233


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
1234 1235 1236
        regularization: A Regularizer, such as
                        fluid.regularizer.L2DecayRegularizer.
        name: A optional name prefix.
1237
    Examples:
Q
qiaolongfei 已提交
1238 1239 1240

      .. code-block:: python

1241
        optimizer = fluid.optimizer.Momentum()
1242 1243
        optimizer.minimize(cost)
        model_average = fluid.optimizer.ModelAverage(0.15,
1244 1245 1246 1247 1248
                                                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()...)
1249 1250 1251 1252

            with model_average.apply(exe):
                for data in test_reader():
                    exe.run(inference_program...)
1253 1254 1255
    """

    def __init__(self,
W
wanghaoshuang 已提交
1256
                 average_window_rate,
1257 1258
                 min_average_window=10000,
                 max_average_window=10000,
X
Xin Pan 已提交
1259 1260 1261 1262
                 regularization=None,
                 name=None):
        super(ModelAverage, self).__init__(
            0.0, regularization=regularization, name=name)
1263 1264 1265
        self.average_window = average_window_rate
        self.min_average_window = min_average_window
        self.max_average_window = max_average_window
1266

1267
        self.params_grads = []
1268 1269
        for param in framework.default_main_program().global_block(
        ).all_parameters():
1270
            if param.do_model_average != False:
1271 1272 1273 1274
                grad = param.block.create_var(
                    name=unique_name.generate(".".join([param.name, 'tmp'])),
                    dtype=param.dtype,
                    persistable=False,
W
wanghaoshuang 已提交
1275
                    stop_gradient=True)
1276
                self.params_grads.append((param, grad))
1277

1278
        for param, grad in self.params_grads:
1279 1280
            if grad is None:
                continue
W
Wu Yi 已提交
1281
            with param.block.program._optimized_guard([param, grad]):
1282
                self._append_average_accumulate_op(param)
1283

1284 1285 1286 1287
        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:
1288
                self._add_average_apply_op(block, param_grad)
1289 1290 1291 1292 1293

        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:
1294
                self._add_average_restore_op(block, param_grad)
1295

1296
    def _add_average_apply_op(self, block, param_grad):
L
Luo Tao 已提交
1297 1298 1299 1300 1301 1302
        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(
1303
            self._get_accumulator('num_accumulates', param))
L
Luo Tao 已提交
1304
        old_num_accumulates = block._clone_variable(
1305
            self._get_accumulator('old_num_accumulates', param))
L
Luo Tao 已提交
1306
        num_updates = block._clone_variable(
1307 1308 1309 1310 1311 1312
            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
1313 1314 1315 1316
        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
1317
        ops._elementwise_div(x=sum, y=tmp, out=param)
1318 1319

    def _add_average_restore_op(self, block, param_grad):
L
Luo Tao 已提交
1320 1321
        param = block._clone_variable(param_grad[0])
        grad = block._clone_variable(param_grad[1])
1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360
        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,
            })

1361 1362
    @contextmanager
    def apply(self, executor, need_restore=True):
1363 1364
        """Apply average values to parameters of current model.
        """
1365 1366 1367 1368 1369 1370
        executor.run(self.apply_program)
        try:
            yield
        finally:
            if need_restore:
                self.restore(executor)
1371 1372 1373 1374

    def restore(self, executor):
        """Restore parameter values of current model.
        """
1375
        executor.run(self.restore_program)