mindspore 2.2.14__cp37-cp37m-manylinux1_x86_64.whl → 2.3.0rc2__cp37-cp37m-manylinux1_x86_64.whl

mindspore/experimental/optim/lr_scheduler.py

@@ -25,7 +25,7 @@ from mindspore.ops import operations as P
 from mindspore import _checkparam as Validator
 
 
-__all__ = ['StepLR', 'LinearLR', 'LRScheduler', 'ExponentialLR', 'PolynomialLR', 'ChainedScheduler',
+__all__ = ['StepLR', 'LinearLR', 'LRScheduler', 'ExponentialLR', 'PolynomialLR',
            'MultiplicativeLR', 'ConstantLR', 'MultiStepLR', 'LambdaLR', 'SequentialLR', 'ReduceLROnPlateau',
            'CyclicLR', 'CosineAnnealingWarmRestarts', 'CosineAnnealingLR']
 
@@ -38,7 +38,7 @@ class LRScheduler:
     .. warning::
         This is an experimental lr scheduler module that is subject to change.
         This module must be used with optimizers in `Experimental Optimizer
-        <https://www.mindspore.cn/docs/en/r2.2/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
+        <https://www.mindspore.cn/docs/en/master/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
 
     Args:
         optimizer (:class:`mindspore.experimental.optim.Optimizer`): The optimizer instance.
@@ -99,7 +99,7 @@ class LRScheduler:
                                    f"in param_groups[{i}] when resuming an optimizer")
         self.base_lrs = [group['initial_lr'] for group in optimizer.param_groups]
         self.optimizer = optimizer
-        self._last_lr = [group['lr'] for group in optimizer.param_groups]
+        self._last_lr = [lr for lr in optimizer.lrs]
         self.groups_num = len(optimizer.param_groups)
         self.last_epoch = Parameter(Tensor(last_epoch, dtype=mstype.float32),
                                     name='last_epoch_' + self.__class__.__name__)
@@ -135,7 +135,7 @@ class LRScheduler:
 
         for i in range(self.groups_num):
             lr = values[i]
-            ops.assign(self.optimizer.param_groups[i]["lr"], lr)
+            ops.assign(self._last_lr[i], lr)
 
         return True
 
@@ -149,7 +149,7 @@ class StepLR(LRScheduler):
     .. warning::
         This is an experimental lr scheduler module that is subject to change.
         This module must be used with optimizers in `Experimental Optimizer
-        <https://www.mindspore.cn/docs/en/r2.2/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
+        <https://www.mindspore.cn/docs/en/master/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
 
     Args:
         optimizer (:class:`mindspore.experimental.optim.Optimizer`): Wrapped optimizer.
@@ -166,7 +166,7 @@ class StepLR(LRScheduler):
         >>> from mindspore import nn
         >>> from mindspore.experimental import optim
         >>> # Define the network structure of LeNet5. Refer to
-        >>> # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/lenet.py
+        >>> # https://gitee.com/mindspore/docs/blob/master/docs/mindspore/code/lenet.py
         >>> net = LeNet5()
         >>> loss_fn = nn.SoftmaxCrossEntropyWithLogits(sparse=True)
         >>> optimizer = optim.Adam(net.trainable_params(), lr=0.05)
@@ -186,22 +186,25 @@ class StepLR(LRScheduler):
         ...     return loss
         >>> for epoch in range(6):
         ...     # Create the dataset taking MNIST as an example. Refer to
-        ...     # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/mnist.py
+        ...     # https://gitee.com/mindspore/docs/blob/master/docs/mindspore/code/mnist.py
         ...     for data, label in create_dataset():
         ...         train_step(data, label)
         ...     scheduler.step()
         ...     current_lr = scheduler.get_last_lr()
     """
     def __init__(self, optimizer, step_size, gamma=0.1, last_epoch=-1):
+        if not isinstance(step_size, int) and not isinstance(step_size, bool):
+            raise TypeError(f"For 'StepLR', the 'step_size' must be int, but got {type(step_size)}.")
+        if not isinstance(gamma, float):
+            raise TypeError(f"For 'StepLR', the 'gamma' must be float, but got {type(gamma)}.")
         self.step_size = step_size
         self.gamma = gamma
         super(StepLR, self).__init__(optimizer, last_epoch)
 
     def get_lr(self):
-        lrs = [lr.value() for lr in self._last_lr]
         if self.last_epoch == 0 or self.last_epoch % self.step_size != 0:
-            return lrs
-        return [lr * self.gamma for lr in lrs]
+            return [lr * 1. for lr in self._last_lr]
+        return [lr * self.gamma for lr in self._last_lr]
 
     def _get_closed_form_lr(self):
         return [base_lr * self.gamma ** (self.last_epoch // self.step_size)
@@ -218,7 +221,7 @@ class LinearLR(LRScheduler):
     .. warning::
         This is an experimental lr scheduler module that is subject to change.
         This module must be used with optimizers in `Experimental Optimizer
-        <https://www.mindspore.cn/docs/en/r2.2/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
+        <https://www.mindspore.cn/docs/en/master/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
 
     Args:
         optimizer (:class:`mindspore.experimental.optim.Optimizer`): Wrapped optimizer.
@@ -243,7 +246,7 @@ class LinearLR(LRScheduler):
         >>> from mindspore import nn
         >>> from mindspore.experimental import optim
         >>> # Define the network structure of LeNet5. Refer to
-        >>> # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/lenet.py
+        >>> # https://gitee.com/mindspore/docs/blob/master/docs/mindspore/code/lenet.py
         >>> net = LeNet5()
         >>> loss_fn = nn.SoftmaxCrossEntropyWithLogits(sparse=True)
         >>> optimizer = optim.Adam(net.trainable_params(), lr=0.05)
@@ -265,7 +268,7 @@ class LinearLR(LRScheduler):
         ...     return loss
         >>> for epoch in range(5):
         ...     # Create the dataset taking MNIST as an example. Refer to
-        ...     # https://gitee.com/mindspore/docs/blob/r2.2/docs/mindspore/code/mnist.py
+        ...     # https://gitee.com/mindspore/docs/blob/master/docs/mindspore/code/mnist.py
         ...     for data, label in create_dataset():
         ...         train_step(data, label)
         ...     scheduler.step()
@@ -279,24 +282,24 @@ class LinearLR(LRScheduler):
 
         if end_factor > 1.0 or end_factor < 0:
             raise ValueError('Ending multiplicative factor expected to be between 0 and 1.')
-
+        if not isinstance(total_iters, int):
+            raise TypeError(f"For 'LinearLR', the type of total_iters must be int, but got {type(total_iters)}.")
         self.start_factor = start_factor
         self.end_factor = end_factor
         self.total_iters = total_iters
         super(LinearLR, self).__init__(optimizer, last_epoch)
 
     def get_lr(self):
-        lrs = [lr.value() for lr in self._last_lr]
 
         if self.last_epoch == 0:
-            return [lr * self.start_factor for lr in lrs]
+            return [lr * self.start_factor for lr in self._last_lr]
 
         if self.last_epoch > self.total_iters:
-            return lrs
+            return [lr * 1. for lr in self._last_lr]
 
         factor = 1. + (self.end_factor - self.start_factor) / (
             self.total_iters * self.start_factor + (self.last_epoch - 1) * (self.end_factor - self.start_factor))
-        return [lr * factor for lr in lrs]
+        return [lr * factor for lr in self._last_lr]
 
     def _get_closed_form_lr(self):
         return [base_lr * (self.start_factor +
@@ -314,7 +317,7 @@ class ExponentialLR(LRScheduler):
     .. warning::
         This is an experimental lr scheduler module that is subject to change.
         This module must be used with optimizers in `Experimental Optimizer
-        <https://www.mindspore.cn/docs/en/r2.2/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
+        <https://www.mindspore.cn/docs/en/master/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
 
     Args:
         optimizer (:class:`mindspore.experimental.optim.Optimizer`): Wrapped optimizer.
@@ -346,14 +349,15 @@ class ExponentialLR(LRScheduler):
     """
 
     def __init__(self, optimizer, gamma, last_epoch=-1):
+        if not isinstance(gamma, float):
+            raise TypeError(f"For 'ExponentialLR', the 'gamma' must be float, but got {type(gamma)}.")
         self.gamma = gamma
         super(ExponentialLR, self).__init__(optimizer, last_epoch)
 
     def get_lr(self):
-        lrs = [lr.value() for lr in self._last_lr]
         if self.last_epoch == 0:
-            return lrs
-        return [lr * self.gamma for lr in lrs]
+            return [lr * 1. for lr in self._last_lr]
+        return [lr * self.gamma for lr in self._last_lr]
 
     def _get_closed_form_lr(self):
         return [base_lr * self.gamma ** self.last_epoch
@@ -380,7 +384,7 @@ class PolynomialLR(LRScheduler):
     .. warning::
         This is an experimental lr scheduler module that is subject to change.
         This module must be used with optimizers in `Experimental Optimizer
-        <https://www.mindspore.cn/docs/en/r2.2/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
+        <https://www.mindspore.cn/docs/en/master/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
 
     Args:
         optimizer (:class:`mindspore.experimental.optim.Optimizer`): Wrapped optimizer.
@@ -416,6 +420,12 @@ class PolynomialLR(LRScheduler):
         [Tensor(shape=[], dtype=Float32, value= 0)]
     """
     def __init__(self, optimizer, total_iters=5, power=1.0, last_epoch=-1):
+        if not isinstance(power, float):
+            raise TypeError(f"For 'PolynomialLR', the 'power' must be float, but got {type(power)}.")
+        if power < 0:
+            raise ValueError(f"For 'PolynomialLR', the 'power' must be >= 0, but got {power}.")
+        if not isinstance(total_iters, int):
+            raise TypeError(f"For 'PolynomialLR', the type of total_iters must be int, but got {type(total_iters)}.")
         self.total_iters = total_iters
         self.power = power
         self.min = P.Minimum()
@@ -423,13 +433,11 @@ class PolynomialLR(LRScheduler):
         super(PolynomialLR, self).__init__(optimizer, last_epoch)
 
     def get_lr(self):
-        lrs = [lr.value() for lr in self._last_lr]
-
         if self.last_epoch == 0 or self.last_epoch > self.total_iters:
-            return lrs
+            return [lr * 1. for lr in self._last_lr]
         factor = ((1.0 - self.last_epoch / self.total_iters) / (
             1.0 - (self.last_epoch - 1) / self.total_iters)) ** self.power
-        return [lr * factor for lr in lrs]
+        return [lr * factor for lr in self._last_lr]
 
     def _get_closed_form_lr(self):
         return [
@@ -437,68 +445,6 @@ class PolynomialLR(LRScheduler):
             for base_lr in self.base_lrs]
 
 
-@jit_class
-class ChainedScheduler:
-    r"""
-    Save the learning rate scheduler chain list of multiple learning rate schedulers,
-    and call the step() function to execute the step() function of each learning rate scheduler.
-
-    .. warning::
-        This is an experimental lr scheduler module that is subject to change.
-        This module must be used with optimizers in `Experimental Optimizer
-        <https://www.mindspore.cn/docs/en/r2.2/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
-
-    Args:
-        schedulers (list[:class:`mindspore.experimental.optim.lr_scheduler.LRScheduler`]):
-            List of learning rate schedulers.
-
-    Supported Platforms:
-        ``Ascend`` ``GPU`` ``CPU``
-
-    Examples:
-        >>> from mindspore import nn
-        >>> from mindspore.experimental import optim
-        >>> class Net(nn.Cell):
-        ...     def __init__(self):
-        ...         super(Net, self).__init__()
-        ...         self.fc = nn.Dense(16 * 5 * 5, 120)
-        ...     def construct(self, x):
-        ...         return self.fc(x)
-        >>> net = Net()
-        >>> optimizer = optim.Adam(net.trainable_params(), 0.01)
-        >>> scheduler1 = optim.lr_scheduler.PolynomialLR(optimizer)
-        >>> scheduler2 = optim.lr_scheduler.ExponentialLR(optimizer, gamma=0.5)
-        >>> scheduler = optim.lr_scheduler.ChainedScheduler([scheduler1, scheduler2])
-        >>> for i in range(6):
-        ...     scheduler.step()
-        ...     current_lr = scheduler.get_last_lr()
-        ...     print(current_lr)
-        [Tensor(shape=[], dtype=Float32, value= 0.004)]
-        [Tensor(shape=[], dtype=Float32, value= 0.0015)]
-        [Tensor(shape=[], dtype=Float32, value= 0.0005)]
-        [Tensor(shape=[], dtype=Float32, value= 0.000125)]
-        [Tensor(shape=[], dtype=Float32, value= 0)]
-        [Tensor(shape=[], dtype=Float32, value= 0)]
-    """
-    def __init__(self, schedulers):
-        self._schedulers = list(schedulers)
-        self.optimizer = schedulers[0].optimizer
-        self._last_lr = [lr for lr in self._schedulers[-1]._last_lr]  # pylint: disable=W0212
-
-    def step(self):
-        """
-        Sequential execution of the saved learning rate scheduler's step() function.
-        """
-        for scheduler in self._schedulers:
-            scheduler.step()
-
-    def get_last_lr(self):
-        """
-        Return last computed learning rate by current scheduler.
-        """
-        return [lr.value() for lr in self._last_lr]
-
-
 @jit_class
 class LambdaLR(LRScheduler):
     """Sets the learning rate of each parameter group to the initial lr
@@ -507,7 +453,7 @@ class LambdaLR(LRScheduler):
     .. warning::
         This is an experimental lr scheduler module that is subject to change.
         This module must be used with optimizers in `Experimental Optimizer
-        <https://www.mindspore.cn/docs/en/r2.2/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
+        <https://www.mindspore.cn/docs/en/master/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
 
     Args:
         optimizer (:class:`mindspore.experimental.optim.Optimizer`): Wrapped optimizer.
@@ -560,7 +506,7 @@ class MultiplicativeLR(LRScheduler):
     .. warning::
         This is an experimental lr scheduler module that is subject to change.
         This module must be used with optimizers in `Experimental Optimizer
-        <https://www.mindspore.cn/docs/en/r2.2/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
+        <https://www.mindspore.cn/docs/en/master/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
 
     Args:
         optimizer (:class:`mindspore.experimental.optim.Optimizer`): Wrapped optimizer.
@@ -598,11 +544,10 @@ class MultiplicativeLR(LRScheduler):
         super(MultiplicativeLR, self).__init__(optimizer, last_epoch)
 
     def get_lr(self):
-        lrs = [lr.value() for lr in self._last_lr]
         if self.last_epoch > 0:
             return [lr * lmbda(self.last_epoch)
-                    for lmbda, lr in zip(self.lr_lambdas, lrs)]
-        return lrs
+                    for lmbda, lr in zip(self.lr_lambdas, self._last_lr)]
+        return [lr * 1. for lr in self._last_lr]
 
 
 @jit_class
@@ -615,7 +560,7 @@ class MultiStepLR(LRScheduler):
     .. warning::
         This is an experimental lr scheduler module that is subject to change.
         This module must be used with optimizers in `Experimental Optimizer
-        <https://www.mindspore.cn/docs/en/r2.2/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
+        <https://www.mindspore.cn/docs/en/master/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
 
     Args:
         optimizer (:class:`mindspore.experimental.optim.Optimizer`): Wrapped optimizer.
@@ -654,11 +599,10 @@ class MultiStepLR(LRScheduler):
         [Tensor(shape=[], dtype=Float32, value= 0.0005)]
         [Tensor(shape=[], dtype=Float32, value= 0.0005)]
     """
-
     def __init__(self, optimizer, milestones, gamma=0.1, last_epoch=-1):
         Validator.check_value_type('milestones', milestones, [list])
         for milestone in milestones:
-            if not isinstance(milestone, int):
+            if not isinstance(milestone, int) and not isinstance(milestone, bool):
                 raise TypeError(f"For 'MultiStepLR', elements of the 'milestones' must be type of int, "
                                 f"but got one element of 'milestones' type: {type(milestone)}.")
         Validator.check_value_type('gamma', gamma, [float, int])
@@ -692,7 +636,7 @@ class ConstantLR(LRScheduler):
     .. warning::
         This is an experimental lr scheduler module that is subject to change.
         This module must be used with optimizers in `Experimental Optimizer
-        <https://www.mindspore.cn/docs/en/r2.2/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
+        <https://www.mindspore.cn/docs/en/master/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
 
     Args:
         optimizer (:class:`mindspore.experimental.optim.Optimizer`): Wrapped optimizer.
@@ -727,18 +671,18 @@ class ConstantLR(LRScheduler):
     def __init__(self, optimizer, factor=1.0 / 3, total_iters=5, last_epoch=-1):
         if factor > 1.0 or factor < 0:
             raise ValueError('Constant multiplicative factor expected to be between 0 and 1.')
-
+        if not isinstance(total_iters, int):
+            raise TypeError(f"For 'ConstantLR', the type of total_iters must be int, but got {type(total_iters)}.")
         self.factor = factor
         self.total_iters = total_iters
         super(ConstantLR, self).__init__(optimizer, last_epoch)
 
     def get_lr(self):
-        lrs = [lr.value() for lr in self._last_lr]
         if self.last_epoch == 0:
-            return [lr * self.factor for lr in lrs]
+            return [lr * self.factor for lr in self._last_lr]
         if self.last_epoch != self.total_iters:
-            return lrs
-        return [lr / self.factor for lr in lrs]
+            return [lr * 1. for lr in self._last_lr]
+        return [lr / self.factor for lr in self._last_lr]
 
     def _get_closed_form_lr(self):
         return [base_lr * (self.factor + (self.last_epoch >= self.total_iters) * (1 - self.factor))
@@ -755,7 +699,7 @@ class SequentialLR:
     .. warning::
         This is an experimental lr scheduler module that is subject to change.
         This module must be used with optimizers in `Experimental Optimizer
-        <https://www.mindspore.cn/docs/en/r2.2/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
+        <https://www.mindspore.cn/docs/en/master/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
 
     Args:
         optimizer (:class:`mindspore.experimental.optim.Optimizer`): Wrapped optimizer.
@@ -812,7 +756,7 @@ class SequentialLR:
         self._schedulers = schedulers
         self.milestones = milestones
         self.milestones_len = len(milestones)
-        self.last_epoch = Parameter(Tensor(last_epoch+1, dtype=mstype.float32),
+        self.last_epoch = Parameter(Tensor(last_epoch + 1, dtype=mstype.float32),
                                     name='last_epoch_' + self.__class__.__name__)
         self.increase_tensor = Tensor(1, mstype.int32)
 
@@ -826,7 +770,6 @@ class SequentialLR:
         self._schedulers[0].step()
         self._last_lr = schedulers[0]._last_lr  # pylint: disable=W0212
 
-
     def step(self):
         """
         Get the current learning rate and change the learning rate.
@@ -860,7 +803,7 @@ class ReduceLROnPlateau:
     .. warning::
         This is an experimental lr scheduler module that is subject to change.
         This module must be used with optimizers in `Experimental Optimizer
-        <https://www.mindspore.cn/docs/en/r2.2/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
+        <https://www.mindspore.cn/docs/en/master/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
 
     Args:
         optimizer (:class:`mindspore.experimental.optim.Optimizer`): Wrapped optimizer.
@@ -986,7 +929,7 @@ class ReduceLROnPlateau:
         else:
             ops.assign_add(self.wait, self.increase_tensor)
 
-        if self.in_cooldown:
+        if self.in_cooldown():
             ops.assign_sub(self.cooldown_counter, self.increase_tensor)
             ops.assign(self.wait, 0)
 
@@ -1005,7 +948,6 @@ class ReduceLROnPlateau:
                 ops.assign(lr, new_lr)
         return True
 
-    @property
     def in_cooldown(self):
         """ Whether in cooldown period. """
         return self.cooldown_counter > 0
@@ -1056,7 +998,7 @@ class CyclicLR(LRScheduler):
     .. warning::
         This is an experimental lr scheduler module that is subject to change.
         This module must be used with optimizers in `Experimental Optimizer
-        <https://www.mindspore.cn/docs/en/r2.2/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
+        <https://www.mindspore.cn/docs/en/master/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
 
     Args:
         optimizer (:class:`mindspore.experimental.optim.Optimizer`): Wrapped optimizer.
@@ -1127,7 +1069,7 @@ class CyclicLR(LRScheduler):
 
         if last_epoch == -1:
             for lr, group in zip(base_lrs, optimizer.param_groups):
-                group['lr'] = Parameter(lr)
+                ops.assign(group['lr'], Parameter(lr))
 
         self.max_lrs = self._preprocess_input_param(optimizer, max_lr, 'max_lr')
         self.max_lrs = [Tensor(lr) for lr in self.max_lrs]
@@ -1233,7 +1175,7 @@ class CosineAnnealingWarmRestarts(LRScheduler):
     .. warning::
         This is an experimental lr scheduler module that is subject to change.
         This module must be used with optimizers in `Experimental Optimizer
-        <https://www.mindspore.cn/docs/en/r2.2/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
+        <https://www.mindspore.cn/docs/en/master/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
 
     Args:
         optimizer (:class:`mindspore.experimental.optim.Optimizer`): Wrapped optimizer.
@@ -1346,7 +1288,7 @@ class CosineAnnealingLR(LRScheduler):
     r"""
     Set the learning rate of each parameter group using a cosine annealing lr
     schedule. Where :math:`\eta_{max}` is set to the initial lr, :math:`\eta_{min}` is the minimum value
-    for learning rate, :math:`\eta_{t}` is the current learning rate, :math:`\T_{max}` is iteration number of cosine
+    for learning rate, :math:`\eta_{t}` is the current learning rate, :math:`T_{max}` is iteration number of cosine
     function, and :math:`T_{cur}` is the number of epochs since the last restart in SGDR.
 
     .. math::
@@ -1360,17 +1302,17 @@ class CosineAnnealingLR(LRScheduler):
         \end{aligned}
 
     For more details, please refer to: `SGDR: Stochastic Gradient Descent with Warm Restarts
-    <https://arxiv.org/abs/1608.03983>`_
+    <https://arxiv.org/abs/1608.03983>`_ .
 
     .. warning::
         This is an experimental lr scheduler module that is subject to change.
         This module must be used with optimizers in `Experimental Optimizer
-        <https://www.mindspore.cn/docs/en/r2.2/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
+        <https://www.mindspore.cn/docs/en/master/api_python/mindspore.experimental.html#experimental-optimizer>`_ .
 
     Args:
         optimizer (:class:`mindspore.experimental.optim.Optimizer`): Wrapped optimizer.
         T_max (int): Maximum number of iterations.
-        eta_min (float, optional): Minimum learning rate. Default: ``0``.
+        eta_min (float, optional): Minimum learning rate. Default: ``0.0``.
         last_epoch (int, optional): The index of the last epoch. Default: ``-1``.
 
     Supported Platforms:
@@ -1394,7 +1336,11 @@ class CosineAnnealingLR(LRScheduler):
         [Tensor(shape=[], dtype=Float32, value= 0.05)]
         [Tensor(shape=[], dtype=Float32, value= 0)]
     """
-    def __init__(self, optimizer, T_max, eta_min=0, last_epoch=-1):
+    def __init__(self, optimizer, T_max, eta_min=0.0, last_epoch=-1):
+        if not isinstance(eta_min, (float, int)):
+            raise TypeError(f"For 'CosineAnnealingLR', the 'eta_min' must be float or int, but got {type(eta_min)}.")
+        if not isinstance(T_max, int) and not isinstance(T_max, bool):
+            raise TypeError(f"For 'CosineAnnealingLR', the 'T_max' must be int, but got {type(eta_min)}.")
         self.T_max = T_max
         self.eta_min = eta_min
         self.math_pi = math.pi
@@ -1403,22 +1349,21 @@ class CosineAnnealingLR(LRScheduler):
         super(CosineAnnealingLR, self).__init__(optimizer, last_epoch)
 
     def get_lr(self):
-        lrs = [lr.value() for lr in self._last_lr]
 
         if self.last_epoch == 0:
-            return lrs
+            return [lr * 1. for lr in self._last_lr]
 
         if (self.last_epoch - 1 - self.T_max) % (2 * self.T_max) == 0:
             pct_pi = self.cast(self.math_pi / self.T_max, mstype.float32)
             return [lr + (base_lr - self.eta_min) *
                     (1 - self.cos(pct_pi)) / 2
                     for base_lr, lr in
-                    zip(self.base_lrs, lrs)]
+                    zip(self.base_lrs, self._last_lr)]
 
         return [(1 + self.cos(self.math_pi * self.last_epoch / self.T_max)) /
                 (1 + self.cos(self.math_pi * (self.last_epoch - 1) / self.T_max)) *
                 (lr - self.eta_min) + self.eta_min
-                for lr in lrs]
+                for lr in self._last_lr]
 
     def _get_closed_form_lr(self):
         return [self.eta_min + (base_lr - self.eta_min) *

mindspore/experimental/optim/nadam.py

@@ -0,0 +1,157 @@
+# Copyright 2023 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.
+# ============================================================================
+"""nadam"""
+from __future__ import absolute_import
+
+from mindspore.ops import functional as F, composite as C, operations as P
+from mindspore.common import Parameter, Tensor
+import mindspore.common.dtype as mstype
+from mindspore import _checkparam as validator
+from mindspore.experimental.optim.optimizer import Optimizer, check_not_less_than, check_not_less_than_without_equal
+from mindspore import jit
+
+_nadam_opt = C.MultitypeFuncGraph("nadam_opt")
+
+op_sqrt = P.Sqrt()
+
+
+@_nadam_opt.register("Number", "Number", "Number", "Number", "Tensor", "Tensor", "Tensor",
+                     "Tensor", "Tensor", "Tensor", "Tensor")
+def _tensor_run_opt(beta1, beta2, momentum_decay, eps, step_t, lr, param, grad, exp_avg, exp_avg_sq, mu_product):
+    """Apply nadam optimizer to the weight parameter."""
+    bias_correction2 = 1 - beta2 ** step_t
+    mu = beta1 * (1. - 0.5 * (0.96 ** (step_t * momentum_decay)))
+    mu_next = beta1 * (1. - 0.5 * (0.96 ** ((step_t + 1) * momentum_decay)))
+    F.assign(mu_product, mu_product * mu)
+    F.assign(exp_avg, exp_avg * beta1 + grad * (1 - beta1))
+    F.assign(exp_avg_sq, exp_avg_sq * beta2 + grad * grad * (1 - beta2))
+
+    denom = op_sqrt(exp_avg_sq / bias_correction2) + eps
+
+    mu_product_next = mu_product * mu_next
+    F.assign(param, param - lr * (1. - mu) / (1. - mu_product) * grad / denom)
+    F.assign(param, param - (lr * mu_next) / (1. - mu_product_next) * exp_avg / denom)
+
+    return True
+
+
+class NAdam(Optimizer):
+    r"""
+    Implements NAdam algorithm.
+
+    .. _Incorporating Nesterov Momentum into Adam:
+        https://openreview.net/forum?id=OM0jvwB8jIp57ZJjtNEZ
+
+    .. warning::
+        This is an experimental optimizer API that is subject to change.
+        This module must be used with lr scheduler module in `LRScheduler Class
+        <https://www.mindspore.cn/docs/en/master/api_python/mindspore.experimental.html#lrscheduler-class>`_ .
+
+    Args:
+        params (Union[list(Parameter), list(dict)]): list of parameters to optimize or dicts defining
+            parameter groups.
+        lr (Union[int, float, Tensor], optional): learning rate. Default: ``2e-3``.
+        betas (Tuple[float, float], optional): coefficients used for computing
+            running averages of gradient and its square. Default: ``(0.9, 0.999)``.
+        eps (float, optional): term added to the denominator to improve
+            numerical stability. Default: ``1e-8``.
+        weight_decay (float, optional): weight decay (L2 penalty). Default: ``0.``.
+        momentum_decay (float, optional): momentum momentum_decay. Default: ``4e-3``.
+
+    Inputs:
+        - **gradients** (tuple[Tensor]) - The gradients of `params`.
+
+    Raises:
+        ValueError: If the learning rate is not int, float or Tensor.
+        ValueError: If the learning rate is less than 0.
+        ValueError: If the `eps` is less than 0.0.
+        ValueError: If the `weight_decay` is less than 0.
+        ValueError: If the `momentum_decay` is less than 0.
+        ValueError: If elements of `betas` not in the range of [0, 1).
+
+    Supported Platforms:
+        ``Ascend`` ``GPU`` ``CPU``
+
+    Examples:
+        >>> import mindspore
+        >>> from mindspore import nn
+        >>> from mindspore.experimental import optim
+        >>> # Define the network structure of LeNet5. Refer to
+        >>> # https://gitee.com/mindspore/docs/blob/master/docs/mindspore/code/lenet.py
+        >>> net = LeNet5()
+        >>> loss_fn = nn.SoftmaxCrossEntropyWithLogits(sparse=True)
+        >>> optimizer = optim.NAdam(net.trainable_params(), lr=0.1)
+        >>> def forward_fn(data, label):
+        ...     logits = net(data)
+        ...     loss = loss_fn(logits, label)
+        ...     return loss, logits
+        >>> grad_fn = mindspore.value_and_grad(forward_fn, None, optimizer.parameters, has_aux=True)
+        >>> def train_step(data, label):
+        ...     (loss, _), grads = grad_fn(data, label)
+        ...     optimizer(grads)
+        ...     return loss
+    """
+
+    def __init__(self, params, lr=2e-3, betas=(0.9, 0.999), eps=1e-8, weight_decay=0.0, momentum_decay=4e-3):
+        check_not_less_than_without_equal(lr, "lr", self.cls_name)
+        check_not_less_than_without_equal(eps, "eps", self.cls_name)
+        check_not_less_than(weight_decay, "weight_decay", self.cls_name)
+        check_not_less_than(momentum_decay, "momentum_decay", self.cls_name)
+
+        validator.check_float_range(betas[0], 0., 1., validator.INC_LEFT, "betas[0]", self.cls_name)
+        validator.check_float_range(betas[1], 0., 1., validator.INC_LEFT, "betas[1]", self.cls_name)
+
+        defaults = dict(lr=lr, betas=betas, eps=eps,
+                        weight_decay=weight_decay, momentum_decay=momentum_decay)
+        super(NAdam, self).__init__(params, defaults)
+        self.step_t = Parameter(Tensor(0, mstype.int32), "step_t")
+        self.exp_avg = self.parameters.clone(prefix="exp_avg", init='zeros')
+        self.exp_avg_sq = self.parameters.clone(prefix="exp_avg_sq", init='zeros')
+        self.mu_product = [Parameter(Tensor(1.), "mu_product_" + param.name) for param in self.parameters]
+
+        self.increase_tensor = Tensor(1, mstype.int32)
+        self.assignadd = P.AssignAdd()
+        self.op_cast = P.Cast()
+
+    @jit
+    def implementation(self, lr, beta1, beta2, weight_decay, momentum_decay, eps, start_id, end_id, gradients):
+        """Extract the common computing part for acceleration"""
+        params = self.parameters[start_id: end_id]
+        grads = gradients[start_id: end_id]
+        grads = self._decay_weight(weight_decay, params, grads)
+        exp_avg = self.exp_avg[start_id: end_id]
+        exp_avg_sq = self.exp_avg_sq[start_id: end_id]
+        mu_product = self.mu_product[start_id: end_id]
+
+        self.hyper_map(F.partial(_nadam_opt, beta1, beta2, momentum_decay, eps, self.step_t, lr),
+                       params, grads, exp_avg, exp_avg_sq, mu_product)
+        return True
+
+    def construct(self, gradients):
+        self.assignadd(self.step_t, self.increase_tensor)
+        for group_id, group in enumerate(self.param_groups):
+
+            lr = self.lrs[group_id]
+            if isinstance(group.get("lr"), float):
+                lr = self.op_cast(group.get("lr"), mstype.float32)
+
+            beta1, beta2 = group["betas"]
+            start_id = self.group_start_id[group_id]
+            end_id = self.group_start_id[group_id + 1]
+            weight_decay = group["weight_decay"]
+            momentum_decay = group["momentum_decay"]
+            eps = group["eps"]
+            self.implementation(lr, beta1, beta2, weight_decay, momentum_decay, eps, start_id, end_id, gradients)
+        return True