!239 Add dynamic learning rate decay and review optimizer code

Merge pull request !239 from fanglei/master

mindspore/nn/dynamic_lr.py

+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# 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
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# 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.
+# ============================================================================
+"""dynamic learning rate"""
+import math
+
+from mindspore._checkparam import ParamValidator as validator
+from mindspore._checkparam import Rel
+
+
+def piecewise_constant_lr(milestone, learning_rates):
+    r"""
+    Get piecewise constant learning rate.
+
+    Calculate learning rate by given `milestone` and `learning_rates`. Let the value of `milestone` be
+    :math:`(M_1, M_2, ..., M_N)` and the value of `learning_rates` be :math:`(x_1, x_2, ..., x_N)`. N is the length of
+    `milestone`. Let the output learning rate be `y`.
+
+    .. math::
+        y[i] = x_t for i \in [M_{t-1}, M_t)
+
+    Args:
+        milestone (list[int]): A list of milestone. This list is a monotone increasing list.
+        learning_rates (list[float]): A list of learning rates.
+
+    Returns:
+        list[float]. The size of list is :math:`M_N`.
+
+    Examples:
+        >>> milestone = [2, 5, 10]
+        >>> learning_rates = [0.1, 0.05, 0.01]
+        >>> lr = piecewise_constant_lr(milestone, learning_rates)
+        [0.1, 0.1, 0.05, 0.05, 0.05, 0.01, 0.01, 0.01, 0.01, 0.01]
+    """
+    validator.check_type('milestone', milestone, (tuple, list))
+    validator.check_type('learning_rates', learning_rates, (tuple, list))
+    if len(milestone) != len(learning_rates):
+        raise ValueError('The size of `milestone` must be same with the size of `learning_rates`.')
+
+    lr = []
+    last_item = 0
+    for i, item in enumerate(milestone):
+        validator.check_integer(f'milestone[{i}]', item, 0, Rel.GT)
+        validator.check_type(f'learning_rates[{i}]', learning_rates[i], [float])
+        if item < last_item:
+            raise ValueError(f'The value of milestone[{i}] must be greater than milestone[{i - 1}]')
+        lr += [learning_rates[i]] * (item - last_item)
+        last_item = item
+
+    return lr
+
+
+def _check_inputs(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch, is_stair):
+    validator.check_integer('total_step', total_step, 0, Rel.GT)
+    validator.check_integer('step_per_epoch', step_per_epoch, 0, Rel.GT)
+    validator.check_integer('decay_epoch', decay_epoch, 0, Rel.GT)
+    validator.check_float_positive('learning_rate', learning_rate)
+    validator.check_float_positive('decay_rate', decay_rate)
+    validator.check_type('is_stair', is_stair, [bool])
+
+
+def exponential_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch, is_stair=False):
+    r"""
+    Calculate learning rate base on exponential decay function.
+
+    For the i-th step, the formula of computing decayed_learning_rate[i] is:
+
+    .. math::
+        decayed\_learning\_rate[i] = learning\_rate * decay\_rate^{\frac{current\_epoch}{decay\_epoch}}
+
+    Where :math:`current\_epoch=floor(\frac{i}{step\_per\_epoch})`.
+
+    Args:
+        learning_rate (float): The initial value of learning rate.
+        decay_rate (float): The decay rate.
+        total_step (int): The total number of steps.
+        step_per_epoch (int): The number of steps in per epoch.
+        decay_epoch (int): A value used to calculate decayed learning rate.
+        is_stair (bool): If true, learning rate decay once every `decay_epoch` times. Default: False.
+
+    Returns:
+        list[float]. The size of list is `total_step`.
+
+    Examples:
+        >>> learning_rate = 0.1
+        >>> decay_rate = 0.9
+        >>> total_step = 6
+        >>> step_per_epoch = 2
+        >>> decay_epoch = 1
+        >>> lr = exponential_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch)
+        [0.1, 0.1, 0.09000000000000001, 0.09000000000000001, 0.08100000000000002, 0.08100000000000002]
+    """
+    _check_inputs(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch, is_stair)
+
+    lr = []
+    for i in range(total_step):
+        if is_stair:
+            lr.append(learning_rate * decay_rate ** math.floor(math.floor(i / step_per_epoch) / decay_epoch))
+        else:
+            lr.append(learning_rate * decay_rate ** (math.floor(i / step_per_epoch) / decay_epoch))
+    return lr
+
+
+def natural_exp_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch, is_stair=False):
+    r"""
+    Calculate learning rate base on natural exponential decay function.
+
+    For the i-th step, the formula of computing decayed_learning_rate[i] is:
+
+    .. math::
+        decayed\_learning\_rate[i] = learning\_rate * e^{-decay\_rate * current\_epoch}
+
+    Where :math:`current\_epoch=floor(\frac{i}{step\_per\_epoch})`.
+
+    Args:
+        learning_rate (float): The initial value of learning rate.
+        decay_rate (float): The decay rate.
+        total_step (int): The total number of steps.
+        step_per_epoch (int): The number of steps in per epoch.
+        decay_epoch (int): A value used to calculate decayed learning rate.
+        is_stair (bool): If true, learning rate decay once every `decay_epoch` times. Default: False.
+
+    Returns:
+        list[float]. The size of list is `total_step`.
+
+    Examples:
+        >>> learning_rate = 0.1
+        >>> decay_rate = 0.9
+        >>> total_step = 6
+        >>> step_per_epoch = 2
+        >>> decay_epoch = 2
+        >>> lr = natural_exp_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch, True)
+        [0.1, 0.1, 0.1, 0.1, 0.016529888822158657, 0.016529888822158657]
+    """
+    _check_inputs(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch, is_stair)
+
+    function = lambda x, y: x
+    if is_stair:
+        function = lambda x, y: math.floor(x / y) * y
+
+    lr = []
+    for i in range(total_step):
+        lr.append(learning_rate * math.e ** (-decay_rate * function(math.floor(i / step_per_epoch), decay_epoch)))
+    return lr
+
+
+def inverse_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch, is_stair=False):
+    r"""
+    Calculate learning rate base on inverse-time decay function.
+
+    For the i-th step, the formula of computing decayed_learning_rate[i] is:
+
+    .. math::
+        decayed\_learning\_rate[i] = learning\_rate / (1 + decay\_rate * current\_epoch / decay\_epoch)
+
+    Where :math:`current\_epoch=floor(\frac{i}{step\_per\_epoch})`.
+
+    Args:
+        learning_rate (float): The initial value of learning rate.
+        decay_rate (float): The decay rate.
+        total_step (int): The total number of steps.
+        step_per_epoch (int): The number of steps in per epoch.
+        decay_epoch (int): A value used to calculate decayed learning rate.
+        is_stair (bool): If true, learning rate decay once every `decay_epoch` times. Default: False.
+
+    Returns:
+        list[float]. The size of list is `total_step`.
+
+    Examples:
+        >>> learning_rate = 0.1
+        >>> decay_rate = 0.5
+        >>> total_step = 6
+        >>> step_per_epoch = 1
+        >>> decay_epoch = 1
+        >>> lr = inverse_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch, True)
+        [0.1, 0.06666666666666667, 0.05, 0.04, 0.03333333333333333, 0.028571428571428574]
+    """
+    _check_inputs(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch, is_stair)
+
+    lr = []
+    for i in range(total_step):
+        if is_stair:
+            lr.append(learning_rate / (1 + decay_rate * math.floor(math.floor(i / step_per_epoch) / decay_epoch)))
+        else:
+            lr.append(learning_rate / (1 + decay_rate * math.floor(i / step_per_epoch) / decay_epoch))
+    return lr
+
+
+def cosine_decay_lr(min_lr, max_lr, total_step, step_per_epoch, decay_epoch):
+    r"""
+    Calculate learning rate base on cosine decay function.
+
+    For the i-th step, the formula of computing decayed_learning_rate[i] is:
+
+    .. math::
+        decayed\_learning\_rate[i] = min\_learning\_rate + 0.5 * (max\_learning\_rate - min\_learning\_rate) *
+        (1 + cos(\frac{current\_epoch}{decay\_epoch}\pi))
+
+    Where :math:`current\_epoch=floor(\frac{i}{step\_per\_epoch})`.
+
+    Args:
+        min_lr (float): The minimum value of learning rate.
+        max_lr (float): The maximum value of learning rate.
+        total_step (int): The total number of steps.
+        step_per_epoch (int): The number of steps in per epoch.
+        decay_epoch (int): A value used to calculate decayed learning rate.
+
+    Returns:
+        list[float]. The size of list is `total_step`.
+
+    Examples:
+        >>> min_lr = 0.01
+        >>> max_lr = 0.1
+        >>> total_step = 6
+        >>> step_per_epoch = 2
+        >>> decay_epoch = 2
+        >>> lr = cosine_decay_lr(min_lr, max_lr, total_step, step_per_epoch, decay_epoch)
+        [0.1, 0.1, 0.05500000000000001, 0.05500000000000001, 0.01, 0.01]
+    """
+    validator.check_float_positive('min_lr', min_lr)
+    validator.check_float_positive('max_lr', max_lr)
+    validator.check_integer('total_step', total_step, 0, Rel.GT)
+    validator.check_integer('step_per_epoch', step_per_epoch, 0, Rel.GT)
+    validator.check_integer('decay_epoch', decay_epoch, 0, Rel.GT)
+
+    delta = 0.5 * (max_lr - min_lr)
+    lr = []
+    for i in range(total_step):
+        tmp_epoch = min(math.floor(i / step_per_epoch), decay_epoch)
+        lr.append(min_lr + delta * (1 + math.cos(math.pi * tmp_epoch / decay_epoch)))
+    return lr
+
+
+def polynomial_decay_lr(learning_rate, end_learning_rate, total_step, step_per_epoch, decay_epoch, power,
+                        update_decay_epoch=False):
+    r"""
+    Calculate learning rate base on polynomial decay function.
+
+    For the i-th step, the formula of computing decayed_learning_rate[i] is:
+
+    .. math::
+        decayed\_learning\_rate[i] = (learning\_rate - end\_learning\_rate) *
+        (1 - tmp\_epoch / decay\_epoch)^{power} + end\_learning\_rate
+
+    Where :math:`tmp\_epoch=min(current\_epoch, decay\_epoch), current\_epoch=floor(\frac{i}{step\_per\_epoch})`.
+    If `update_decay_epoch` is true, update the value of `decay_epoch` every epoch. The formula is
+    :math:`decay\_epoch = decay\_epoch * ceil(current\_epoch / decay\_epoch)`
+
+    Args:
+        learning_rate (float): The initial value of learning rate.
+        end_learning_rate (float): The end value of learning rate.
+        total_step (int): The total number of steps.
+        step_per_epoch (int): The number of steps in per epoch.
+        decay_epoch (int): A value used to calculate decayed learning rate.
+        power (float): A value used to calculate decayed learning rate.
+        update_decay_epoch (bool): If true, update `decay_epoch`. Default: False.
+
+    Returns:
+        list[float]. The size of list is `total_step`.
+
+    Examples:
+        >>> learning_rate = 0.1
+        >>> end_learning_rate = 0.01
+        >>> total_step = 6
+        >>> step_per_epoch = 2
+        >>> decay_epoch = 2
+        >>> power = 0.5
+        >>> lr = polynomial_decay_lr(learning_rate, end_learning_rate, total_step, step_per_epoch, decay_epoch, power)
+        [0.1, 0.1, 0.07363961030678928, 0.07363961030678928, 0.01, 0.01]
+    """
+    validator.check_float_positive('learning_rate', learning_rate)
+    validator.check_float_positive('end_learning_rate', end_learning_rate)
+    validator.check_integer('total_step', total_step, 0, Rel.GT)
+    validator.check_integer('step_per_epoch', step_per_epoch, 0, Rel.GT)
+    validator.check_integer('decay_epoch', decay_epoch, 0, Rel.GT)
+    validator.check_type('power', power, [float])
+    validator.check_type('update_decay_epoch', update_decay_epoch, [bool])
+
+    function = lambda x, y: (x, min(x, y))
+    if update_decay_epoch:
+        function = lambda x, y: (x * max(math.ceil(y / x), 1), y)
+
+    lr = []
+    delta = learning_rate - end_learning_rate
+    for i in range(total_step):
+        current_epoch = math.floor(i / step_per_epoch)
+        decay_epoch, tmp_epoch = function(decay_epoch, current_epoch)
+        lr.append(delta * (1 - tmp_epoch / decay_epoch) ** power + end_learning_rate)
+    return lr

mindspore/nn/optim/adam.py

@@ -13,7 +13,6 @@
 # limitations under the License.
 # ============================================================================
 """adam"""
-from typing import Iterable
 import numpy as np
 
 from mindspore.common import dtype as mstype
@@ -25,7 +24,7 @@ from mindspore.common.parameter import Parameter
 from mindspore.common.tensor import Tensor
 from mindspore._checkparam import ParamValidator as validator
 from mindspore._checkparam import Rel
-from .optimizer import Optimizer, apply_decay, grad_scale
+from .optimizer import Optimizer
 
 _learning_rate_update_func = ['linear', 'cos', 'sin']
 
@@ -168,22 +167,13 @@ class Adam(Optimizer):
     def __init__(self, params, learning_rate=1e-3, beta1=0.9, beta2=0.999, eps=1e-8, use_locking=False,
                  use_nesterov=False, weight_decay=0.0, loss_scale=1.0,
                  decay_filter=lambda x: 'beta' not in x.name and 'gamma' not in x.name):
-        super(Adam, self).__init__(learning_rate, params)
+        super(Adam, self).__init__(learning_rate, params, weight_decay, loss_scale, decay_filter)
         _check_param_value(beta1, beta2, eps, weight_decay)
         validator.check_type("use_locking", use_locking, [bool])
         validator.check_type("use_nesterov", use_nesterov, [bool])
         validator.check_type("loss_scale", loss_scale, [float])
         validator.check_number_range("loss_scale", loss_scale, 1.0, float("inf"), Rel.INC_LEFT)
 
-        self.dynamic_lr = False
-        if isinstance(learning_rate, Iterable) or \
-                (isinstance(learning_rate, Tensor) and learning_rate.dim() == 1):
-            self.dynamic_lr = True
-            self.gather = P.GatherV2()
-            self.assignadd = P.AssignAdd()
-            self.global_step = Parameter(initializer(0, [1], mstype.int32), name="global_step")
-            self.axis = 0
-
         self.beta1 = Tensor(beta1, mstype.float32)
         self.beta2 = Tensor(beta2, mstype.float32)
         self.beta1_power = Parameter(initializer(1, [1], mstype.float32), name="beta1_power")
@@ -196,8 +186,6 @@ class Adam(Optimizer):
         self.decay_tf = tuple(decay_filter(x) for x in self.parameters)
         self.hyper_map = C.HyperMap()
         self.opt = P.Adam(use_locking, use_nesterov)
-        self.weight_decay = weight_decay * loss_scale
-        self.reciprocal_scale = 1.0 / loss_scale
 
         self.pow = P.Pow()
         self.sqrt = P.Sqrt()
@@ -208,15 +196,9 @@ class Adam(Optimizer):
         params = self.parameters
         moment1 = self.moment1
         moment2 = self.moment2
-        if self.weight_decay > 0:
-            gradients = self.hyper_map(F.partial(apply_decay, self.weight_decay), self.decay_tf, params, gradients)
-        if self.reciprocal_scale != 1.0:
-            gradients = self.hyper_map(F.partial(grad_scale, self.reciprocal_scale), gradients)
-
-        lr = self.learning_rate
-        if self.dynamic_lr:
-            lr = self.gather(self.learning_rate, self.global_step, self.axis)
-            F.control_depend(lr, self.assignadd(self.global_step, self.one))
+        gradients = self.decay_weight(gradients)
+        gradients = self.scale_grad(gradients)
+        lr = self.get_lr()
 
         beta1_power = self.beta1_power * self.beta1
         self.beta1_power = beta1_power

mindspore/nn/optim/momentum.py

@@ -13,14 +13,9 @@
 # limitations under the License.
 # ============================================================================
 """momentum"""
-from typing import Iterable
-
 from mindspore.ops import functional as F, composite as C, operations as P
-from mindspore.common.initializer import initializer
 from mindspore.common.parameter import Parameter
-import mindspore.common.dtype as mstype
-from mindspore.common import Tensor
-from .optimizer import Optimizer, apply_decay, grad_scale
+from .optimizer import Optimizer
 
 momentum_opt = C.MultitypeFuncGraph("momentum_opt")
 
@@ -88,43 +83,20 @@ class Momentum(Optimizer):
     """
     def __init__(self, params, learning_rate, momentum, weight_decay=0.0, loss_scale=1.0,
                  decay_filter=lambda x: 'beta' not in x.name and 'gamma' not in x.name):
-        super(Momentum, self).__init__(learning_rate, params)
+        super(Momentum, self).__init__(learning_rate, params, weight_decay, loss_scale, decay_filter)
         if isinstance(momentum, float) and momentum < 0.0:
             raise ValueError("momentum should be at least 0.0, but got momentum {}".format(momentum))
-        if isinstance(learning_rate, Iterable) or \
-                (isinstance(learning_rate, Tensor) and learning_rate.dim() == 1):
-            self.dynamic_lr = True
-            self.gather = P.GatherV2()
-            self.assignadd = P.AssignAdd()
-            self.global_step = Parameter(initializer(0, [1], mstype.int32), name="global_step")
-            self.axis = 0
-        else:
-            self.dynamic_lr = False
-            self.gather = None
-            self.assignadd = None
-            self.global_step = None
-            self.axis = None
         self.momentum = Parameter(momentum, name="momentum")
         self.params = self.parameters
         self.moments = self.params.clone(prefix="moments", init='zeros')
-        self.decay_tf = tuple(decay_filter(x) for x in self.parameters)
         self.hyper_map = C.HyperMap()
         self.opt = P.ApplyMomentum()
-        self.weight_decay = weight_decay * loss_scale
-        self.reciprocal_scale = 1.0 / loss_scale
-        self.one = Tensor(1, mstype.int32)
 
     def construct(self, gradients):
         params = self.params
         moments = self.moments
-        if self.weight_decay > 0:
-            gradients = self.hyper_map(F.partial(apply_decay, self.weight_decay), self.decay_tf, params, gradients)
-        if self.reciprocal_scale != 1.0:
-            gradients = self.hyper_map(F.partial(grad_scale, self.reciprocal_scale), gradients)
-        if self.dynamic_lr:
-            lr = self.gather(self.learning_rate, self.global_step, self.axis)
-            F.control_depend(lr, self.assignadd(self.global_step, self.one))
-        else:
-            lr = self.learning_rate
+        gradients = self.decay_weight(gradients)
+        gradients = self.scale_grad(gradients)
+        lr = self.get_lr()
         success = self.hyper_map(F.partial(momentum_opt, self.opt, lr, self.momentum), gradients, params, moments)
         return success

mindspore/nn/optim/optimizer.py

@@ -17,9 +17,11 @@ from typing import Iterable
 
 import numpy as np
 
+import mindspore
 from mindspore.ops import functional as F, composite as C, operations as P
 from mindspore.nn.cell import Cell
 from mindspore.common.parameter import Parameter, ParameterTuple
+from mindspore.common.initializer import initializer
 from mindspore._checkparam import ParamValidator as validator
 from mindspore._checkparam import Rel
 from mindspore.common.tensor import Tensor
@@ -42,34 +44,110 @@ class Optimizer(Cell):
     Args:
         learning_rate (float): A floating point value for the learning rate. Should be greater than 0.
         parameters (list): A list of parameter, which will be updated. The element in `parameters`
-                           should be class mindspore.Parameter.
+            should be class mindspore.Parameter.
+        weight_decay (float): A floating point value for the weight decay. Default: 0.0.
+        loss_scale (float): A floating point value for the loss scale. Default: 1.0. Should be greater than 0.
+        decay_filter (Function): A function to determine whether to apply weight decay on parameters. Default: lambda
+            x: 'beta' not in x.name and 'gamma' not in x.name.
 
     Raises:
         ValueError: If the learning_rate is a Tensor, but the dims of tensor is greater than 1.
         TypeError: If the learning_rate is not any of the three types: float, Tensor, Iterable.
     """
 
-    def __init__(self, learning_rate, parameters):
+    def __init__(self, learning_rate, parameters, weight_decay=0.0, loss_scale=1.0,
+                 decay_filter=lambda x: 'beta' not in x.name and 'gamma' not in x.name):
         super(Optimizer, self).__init__()
         if isinstance(learning_rate, float):
+            self.dynamic_lr = False
+            self.gather = None
+            self.assignadd = None
+            self.global_step = None
             validator.check_number_range("learning rate", learning_rate, 0.0, float("inf"), Rel.INC_LEFT)
-        elif isinstance(learning_rate, Iterable):
-            learning_rate = Tensor(np.array(list(learning_rate)).astype(np.float32))
-        elif isinstance(learning_rate, Tensor):
-            if learning_rate.dim() > 1:
-                raise ValueError("Learning rate should be a 0 or 1 dim `Tensor`,"
-                                 f"but got {learning_rate.dim()}.")
         else:
-            raise TypeError("Learning rate should be float, Tensor or Iterable.")
+            self.dynamic_lr = True
+            self.gather = P.GatherV2()
+            self.assignadd = P.AssignAdd()
+            self.global_step = Parameter(initializer(0, [1], mindspore.int32), name='global_step')
+            if isinstance(learning_rate, Iterable):
+                learning_rate = Tensor(np.array(list(learning_rate)).astype(np.float32))
+            elif isinstance(learning_rate, Tensor):
+                if learning_rate.dim() > 1:
+                    raise ValueError("Learning rate should be a 0 or 1 dim `Tensor`,"
+                                     f"but got {learning_rate.dim()}.")
+                if learning_rate.dim() == 1 and learning_rate.size() < 2:
+                    logger.warning("If want to use the dynamic learning rate, please make sure that the number "
+                                   "of elements in the list, tuple or tensor passed is greater than 1.")
+            else:
+                raise TypeError("Learning rate should be float, Tensor or Iterable.")
+
+        if loss_scale <= 0.0:
+            raise ValueError("Loss scale should be greater than 0, but got {}".format(loss_scale))
+        if weight_decay < 0.0:
+            raise ValueError("Weight decay should be equal or greater than 0, but got {}".format(weight_decay))
 
-        if isinstance(learning_rate, Tensor) and learning_rate.dim() == 1 and learning_rate.size() < 2:
-            logger.warning("If want to use the dynamic learning rate, please make sure that "
-                           "the number of elements in the list, tuple or tensor passed is greater than 1.")
         self.learning_rate = Parameter(learning_rate, name="learning_rate")
         self.parameters = ParameterTuple(parameters)
+        self.reciprocal_scale = 1.0 / loss_scale
+        self.weight_decay = weight_decay * loss_scale
+        self.decay_flags = tuple(decay_filter(x) for x in self.parameters)
+
         if not self.parameters:
             raise ValueError("optimizer got an empty parameter list.")
 
+    def decay_weight(self, gradients):
+        """
+        Weight decay.
+
+        An approach to reduce the overfitting of a deep learning neural network model.
+
+        Args:
+            gradients (tuple[Tensor]): The gradients of `self.parameters`, and have the same shape with
+                `self.parameters`.
+
+        Returns:
+            tuple[Tensor], The gradients after weight decay.
+        """
+        if self.weight_decay > 0:
+            params = self.params
+            gradients = self.hyper_map(F.partial(apply_decay, self.weight_decay), self.decay_flags, params, gradients)
+
+        return gradients
+
+    def scale_grad(self, gradients):
+        """
+        Loss scale for mixed precision.
+
+        An approach of mixed precision training to improve the speed and energy efficiency of training deep neural
+        network.
+
+        Args:
+            gradients (tuple[Tensor]): The gradients of `self.parameters`, and have the same shape with
+                `self.parameters`.
+
+        Returns:
+            tuple[Tensor], The gradients after loss scale.
+
+        """
+        if self.reciprocal_scale != 1.0:
+            gradients = self.hyper_map(F.partial(grad_scale, self.reciprocal_scale), gradients)
+
+        return gradients
+
+    def get_lr(self):
+        """
+        Get the learning rate of current step.
+
+        Returns:
+            float, the learning rate of current step.
+        """
+        lr = self.learning_rate
+        if self.dynamic_lr:
+            lr = self.gather(self.learning_rate, self.global_step, 0)
+            F.control_depend(lr, self.assignadd(self.global_step, 1))
+
+        return lr
+
     def construct(self, *hyper_params):
         raise NotImplementedError
 

mindspore/nn/optim/rmsprop.py

@@ -14,12 +14,8 @@
 # ============================================================================
 """rmsprop"""
 from mindspore.ops import functional as F, composite as C, operations as P
-from mindspore.common.initializer import initializer
-from mindspore.common.parameter import Parameter
 from mindspore._checkparam import ParamValidator as validator
-import mindspore.common.dtype as mstype
-from mindspore.common import Tensor
-from .optimizer import Optimizer, grad_scale, apply_decay
+from .optimizer import Optimizer
 
 rmsprop_opt = C.MultitypeFuncGraph("rmsprop_opt")
 centered_rmsprop_opt = C.MultitypeFuncGraph("rmsprop_opt")
@@ -138,7 +134,7 @@ class RMSProp(Optimizer):
     def __init__(self, params, learning_rate=0.1, decay=0.9, momentum=0.0, epsilon=1e-10,
                  use_locking=False, centered=False, loss_scale=1.0, weight_decay=0.0,
                  decay_filter=lambda x: 'beta' not in x.name and 'gamma' not in x.name):
-        super(RMSProp, self).__init__(learning_rate, params)
+        super(RMSProp, self).__init__(learning_rate, params, weight_decay, loss_scale, decay_filter)
 
         if isinstance(momentum, float) and momentum < 0.0:
             raise ValueError("momentum should be at least 0.0, but got momentum {}".format(momentum))
@@ -157,15 +153,6 @@ class RMSProp(Optimizer):
         else:
             self.opt = P.ApplyRMSProp(use_locking)
 
-        self.dynamic_lr = False
-        if not isinstance(learning_rate, float):
-            self.dynamic_lr = True
-            self.gather = P.GatherV2()
-            self.assignadd = P.AssignAdd()
-            self.global_step = Parameter(initializer(0, [1], mstype.int32), name="global_step")
-            self.axis = 0
-            self.one = Tensor(1, mstype.int32)
-
         self.momentum = momentum
 
         self.ms = self.parameters.clone(prefix="mean_square", init='zeros')
@@ -173,21 +160,12 @@ class RMSProp(Optimizer):
         self.hyper_map = C.HyperMap()
 
         self.decay = decay
-        self.decay_tf = tuple(decay_filter(x) for x in self.parameters)
-        self.reciprocal_scale = 1.0 / loss_scale
-        self.weight_decay = weight_decay * loss_scale
 
     def construct(self, gradients):
         params = self.parameters
-        if self.weight_decay > 0:
-            gradients = self.hyper_map(F.partial(apply_decay, self.weight_decay), self.decay_tf, params, gradients)
-        if self.reciprocal_scale != 1.0:
-            gradients = self.hyper_map(F.partial(grad_scale, self.reciprocal_scale), gradients)
-        if self.dynamic_lr:
-            lr = self.gather(self.learning_rate, self.global_step, self.axis)
-            F.control_depend(lr, self.assignadd(self.global_step, self.one))
-        else:
-            lr = self.learning_rate
+        gradients = self.decay_weight(gradients)
+        gradients = self.scale_grad(gradients)
+        lr = self.get_lr()
         if self.centered:
             success = self.hyper_map(F.partial(centered_rmsprop_opt, self.opt, lr, self.decay, self.epsilon,
                                                self.momentum), params, self.mg, self.ms, self.moment, gradients)

mindspore/nn/optim/sgd.py

@@ -14,11 +14,9 @@
 # ============================================================================
 """sgd"""
 from mindspore.ops import functional as F, composite as C, operations as P
-from mindspore.common.initializer import initializer
 from mindspore.common.parameter import Parameter
 from mindspore._checkparam import ParamValidator as validator
-import mindspore.common.dtype as mstype
-from .optimizer import Optimizer, grad_scale
+from .optimizer import Optimizer
 
 sgd_opt = C.MultitypeFuncGraph("sgd_opt")
 
@@ -83,7 +81,7 @@ class SGD(Optimizer):
     def __init__(self, params, learning_rate=0.1, momentum=0.0, dampening=0.0, weight_decay=0.0, nesterov=False,
                  loss_scale=1.0):
 
-        super(SGD, self).__init__(learning_rate, params)
+        super(SGD, self).__init__(learning_rate, params, weight_decay, loss_scale)
 
         if isinstance(momentum, float) and momentum < 0.0:
             raise ValueError("momentum should be at least 0.0, but got momentum {}".format(momentum))
@@ -92,44 +90,22 @@ class SGD(Optimizer):
             raise ValueError("dampening should be at least 0.0, but got dampening {}".format(dampening))
         self.dampening = dampening
 
-        if weight_decay < 0.0:
-            raise ValueError("weight_decay should be at least 0.0, but got weight_decay {}".format(weight_decay))
-        self.weight_decay = weight_decay
-
         validator.check_type("nesterov", nesterov, [bool])
         self.nesterov = nesterov
 
         self.opt = P.SGD(dampening, weight_decay, nesterov)
 
-        self.dynamic_lr = False
-        self.gather = None
-        self.global_step = None
-        self.axis = None
-        if not isinstance(learning_rate, float):
-            self.dynamic_lr = True
-            self.gather = P.GatherV2()
-            self.assignadd = P.AssignAdd()
-            self.global_step = Parameter(initializer(0, [1], mstype.int32), name="global_step")
-            self.axis = 0
         self.momentum = Parameter(momentum, name="momentum")
-        self.params = self.parameters
-        self.accum = self.params.clone(prefix="accum", init='zeros')
-        self.stat = self.params.clone(prefix="stat", init='ones')
+        self.accum = self.parameters.clone(prefix="accum", init='zeros')
+        self.stat = self.parameters.clone(prefix="stat", init='ones')
         self.hyper_map = C.HyperMap()
 
-        self.weight_decay = weight_decay * loss_scale
-        self.reciprocal_scale = 1.0 / loss_scale
-
     def construct(self, gradients):
-        params = self.params
+        params = self.parameters
         accum = self.accum
         stat = self.stat
-        if self.reciprocal_scale != 1.0:
-            gradients = self.hyper_map(F.partial(grad_scale, self.reciprocal_scale), gradients)
-        if self.dynamic_lr:
-            lr = self.gather(self.learning_rate, self.global_step, self.axis)
-            F.control_depend(lr, self.assignadd(self.global_step, 1))
-        else:
-            lr = self.learning_rate
+        gradients = self.decay_weight(gradients)
+        gradients = self.scale_grad(gradients)
+        lr = self.get_lr()
         success = self.hyper_map(F.partial(sgd_opt, self.opt, lr, self.momentum), gradients, params, accum, stat)
         return success

tests/ut/python/nn/optim/test_optimizer.py

@@ -15,17 +15,11 @@
 """ test optimizer """
 import numpy as np
 import pytest
-from mindspore.nn.optim import Optimizer, SGD, Adam, AdamWeightDecay, AdamWeightDecayDynamicLR
 from mindspore import Tensor
+from mindspore.nn.optim import Optimizer, SGD, Adam, AdamWeightDecay, AdamWeightDecayDynamicLR
 from mindspore.common.parameter import Parameter
 
 
-gradient = Tensor(np.zeros([1, 2, 3]))
-accumulation = gradient
-variable = accumulation
-
-
-paramsTensor = Tensor(np.zeros([1, 2, 3]))
 class IterableObjc:
     def __iter__(self):
         cont = 0
@@ -56,6 +50,7 @@ class TestAdam():
 
     def test_construct(self):
         with pytest.raises(TypeError):
+            gradient = Tensor(np.zeros([1, 2, 3]))
             adam = Adam(params, learning_rate=1e-3, beta1=0.9, beta2=0.999, eps=1e-8, use_locking=False,
                         use_nesterov=False, weight_decay=0.0, loss_scale=1.0)
             adam.construct(gradient)
@@ -105,4 +100,5 @@ class TestUnsupportParam():
 
     def test_Sgd_init(self):
         with pytest.raises(TypeError):
+            paramsTensor = Tensor(np.zeros([1, 2, 3]))
             SGD(paramsTensor)

tests/ut/python/nn/test_dynamic_lr.py

+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# 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
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# 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.
+# ============================================================================
+""" Test Dynamic Learning Rate """
+import pytest
+import mindspore
+from mindspore.nn import dynamic_lr as dr
+
+milestone = [10, 20, 30]
+learning_rates = [0.1, 0.05, 0.01]
+learning_rate = 0.1
+end_learning_rate = 0.01
+decay_rate = 0.9
+total_step = 30
+step_per_epoch = 3
+decay_epoch = 2
+min_lr = 0.01
+max_lr = 0.1
+power = 0.5
+
+class TestInputs:
+    def test_milestone1(self):        
+        milestone1 = 1
+        with pytest.raises(ValueError):
+            dr.piecewise_constant_lr(milestone1, learning_rates)
+
+    def test_milestone2(self):
+        milestone1 = [20, 10, 1]
+        with pytest.raises(ValueError):
+            dr.piecewise_constant_lr(milestone1, learning_rates)
+
+        milestone2 = [1.0, 2.0, True]
+        with pytest.raises(ValueError):
+            dr.piecewise_constant_lr(milestone2, learning_rates)
+
+    def test_learning_rates1(self):
+        lr = True
+        with pytest.raises(ValueError):
+            dr.piecewise_constant_lr(milestone, lr)
+
+    def test_learning_rates2(self):
+        lr = [1, 2, 1]
+        with pytest.raises(ValueError):
+            dr.piecewise_constant_lr(milestone, lr)
+
+    def test_learning_rate_type(self):
+        lr = True
+        with pytest.raises(TypeError):
+            dr.exponential_decay_lr(lr, decay_rate, total_step, step_per_epoch, decay_epoch)
+        
+        with pytest.raises(TypeError):
+            dr.polynomial_decay_lr(lr, end_learning_rate, total_step, step_per_epoch, decay_epoch, power)
+
+    def test_learning_rate_value(self):
+        lr = -1.0
+        with pytest.raises(ValueError):
+            dr.exponential_decay_lr(lr, decay_rate, total_step, step_per_epoch, decay_epoch)
+
+        with pytest.raises(ValueError):
+            dr.polynomial_decay_lr(lr, end_learning_rate, total_step, step_per_epoch, decay_epoch, power)
+
+    def test_end_learning_rate_type(self):
+        lr = True        
+        with pytest.raises(TypeError):
+            dr.polynomial_decay_lr(learning_rate, lr, total_step, step_per_epoch, decay_epoch, power)
+
+    def test_end_learning_rate_value(self):
+        lr = -1.0
+        with pytest.raises(ValueError):
+            dr.polynomial_decay_lr(learning_rate, lr, total_step, step_per_epoch, decay_epoch, power)
+
+    def test_decay_rate_type(self):
+        rate = 'a'
+        with pytest.raises(TypeError):
+            dr.exponential_decay_lr(learning_rate, rate, total_step, step_per_epoch, decay_epoch)
+
+    def test_decay_rate_value(self):
+        rate = -1.0
+        with pytest.raises(ValueError):
+            dr.exponential_decay_lr(learning_rate, rate, total_step, step_per_epoch, decay_epoch)
+
+    def test_total_step1(self):
+        total_step1 = 2.0
+        with pytest.raises(ValueError):
+            dr.exponential_decay_lr(learning_rate, decay_rate, total_step1, step_per_epoch, decay_epoch)
+
+        with pytest.raises(ValueError):
+            dr.cosine_decay_lr(min_lr, max_lr, total_step1, step_per_epoch, decay_epoch)
+
+        with pytest.raises(ValueError):
+            dr.polynomial_decay_lr(learning_rate, end_learning_rate, total_step1, step_per_epoch, decay_epoch, power)
+
+    def test_total_step2(self):
+        total_step1 = -1
+        with pytest.raises(ValueError):
+            dr.exponential_decay_lr(learning_rate, decay_rate, total_step1, step_per_epoch, decay_epoch)
+
+        with pytest.raises(ValueError):
+            dr.cosine_decay_lr(min_lr, max_lr, total_step1, step_per_epoch, decay_epoch)
+
+        with pytest.raises(ValueError):
+            dr.polynomial_decay_lr(learning_rate, end_learning_rate, total_step1, step_per_epoch, decay_epoch, power)
+
+    def test_step_per_epoch1(self):
+        step_per_epoch1 = True
+        with pytest.raises(ValueError):
+            dr.exponential_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch1, decay_epoch)
+
+        with pytest.raises(ValueError):
+            dr.cosine_decay_lr(min_lr, max_lr, total_step, step_per_epoch1, decay_epoch)
+
+        with pytest.raises(ValueError):
+            dr.polynomial_decay_lr(learning_rate, end_learning_rate, total_step, step_per_epoch1, decay_epoch, power)
+
+    def test_step_per_epoch2(self):
+        step_per_epoch1 = -1
+        with pytest.raises(ValueError):
+            dr.exponential_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch1, decay_epoch)
+        
+        with pytest.raises(ValueError):
+            dr.cosine_decay_lr(min_lr, max_lr, total_step, step_per_epoch1, decay_epoch)
+
+        with pytest.raises(ValueError):
+            dr.polynomial_decay_lr(learning_rate, end_learning_rate, total_step, step_per_epoch1, decay_epoch, power)
+
+    def test_decay_epoch1(self):
+        decay_epoch1 = 'm'
+        with pytest.raises(ValueError):
+            dr.exponential_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch1)
+
+        with pytest.raises(ValueError):
+            dr.cosine_decay_lr(min_lr, max_lr, total_step, step_per_epoch, decay_epoch1)
+
+        with pytest.raises(ValueError):
+            dr.polynomial_decay_lr(learning_rate, end_learning_rate, total_step, step_per_epoch, decay_epoch1, power)
+
+    def test_decay_epoch2(self):
+        decay_epoch1 = -1
+        with pytest.raises(ValueError):
+            dr.exponential_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch1)
+
+        with pytest.raises(ValueError):
+            dr.cosine_decay_lr(min_lr, max_lr, total_step, step_per_epoch, decay_epoch1)
+
+        with pytest.raises(ValueError):
+            dr.polynomial_decay_lr(learning_rate, end_learning_rate, total_step, step_per_epoch, decay_epoch1, power)
+
+    def test_is_stair(self):
+        is_stair = 1
+        with pytest.raises(ValueError):
+            dr.exponential_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch, is_stair)
+
+    def test_min_lr_type(self):
+        min_lr1 = True
+        with pytest.raises(TypeError):
+            dr.cosine_decay_lr(min_lr1, max_lr, total_step, step_per_epoch, decay_epoch)
+
+    def test_min_lr_value(self):
+        min_lr1 = -1.0
+        with pytest.raises(ValueError):
+            dr.cosine_decay_lr(min_lr1, max_lr, total_step, step_per_epoch, decay_epoch)
+
+    def test_max_lr_type(self):
+        max_lr1 = 'a'
+        with pytest.raises(TypeError):
+            dr.cosine_decay_lr(min_lr, max_lr1, total_step, step_per_epoch, decay_epoch)
+
+    def test_max_lr_value(self):
+        max_lr1 = -1.0
+        with pytest.raises(ValueError):
+            dr.cosine_decay_lr(min_lr, max_lr1, total_step, step_per_epoch, decay_epoch)
+
+    def test_power(self):
+        power1 = True
+        with pytest.raises(ValueError):
+            dr.polynomial_decay_lr(learning_rate, end_learning_rate, total_step, step_per_epoch, decay_epoch, power1)
+
+    def test_update_decay_epoch(self):
+        update_decay_epoch = 1
+        with pytest.raises(ValueError):
+            dr.polynomial_decay_lr(learning_rate, end_learning_rate, total_step, step_per_epoch, decay_epoch,
+                                   power, update_decay_epoch)
+
+
+def test_learning_rate():
+    lr = dr.piecewise_constant_lr(milestone, learning_rates)
+    assert len(lr) == milestone[-1]
+
+
+def test_exponential_decay():
+    lr1 = dr.exponential_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch)
+    assert len(lr1) == total_step
+
+    lr2 = dr.exponential_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch, True)
+    assert len(lr2) == total_step
+
+
+def test_enatural_exp_decay():
+    lr1 = dr.natural_exp_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch)
+    assert len(lr1) == total_step
+
+    lr2 = dr.natural_exp_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch, True)
+    assert len(lr2) == total_step
+
+
+def test_inverse_decay():
+    lr1 = dr.inverse_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch)
+    assert len(lr1) == total_step
+
+    lr2 = dr.inverse_decay_lr(learning_rate, decay_rate, total_step, step_per_epoch, decay_epoch, True)
+    assert len(lr2) == total_step
+
+
+def test_cosine_decay():
+    lr = dr.cosine_decay_lr(min_lr, max_lr, total_step, step_per_epoch, decay_epoch)
+    assert len(lr) == total_step
+
+def test_polynomial_decay():
+    lr1 = dr.polynomial_decay_lr(learning_rate, end_learning_rate, total_step, step_per_epoch, decay_epoch, power)
+    assert len(lr1) == total_step
+    lr2 = dr.polynomial_decay_lr(learning_rate, end_learning_rate, total_step, step_per_epoch, decay_epoch, power,
+                                True)
+    assert len(lr2) == total_step