forked from mindspore-Ecosystem/mindspore
66 lines
2.2 KiB
ReStructuredText
66 lines
2.2 KiB
ReStructuredText
mindspore.nn.PolynomialDecayLR
|
||
====================================
|
||
|
||
.. py:class:: mindspore.nn.PolynomialDecayLR(learning_rate, end_learning_rate, decay_steps, power, update_decay_steps=False)
|
||
|
||
基于多项式衰减函数计算学习率。
|
||
|
||
对于当前step,计算decayed_learning_rate[current_step]的公式为:
|
||
|
||
.. math::
|
||
decayed\_learning\_rate[current\_step] = &(learning\_rate - end\_learning\_rate) *\\
|
||
&(1 - tmp\_step / tmp\_decay\_steps)^{power}\\
|
||
&+ end\_learning\_rate
|
||
|
||
其中,
|
||
|
||
.. math::
|
||
tmp\_step=min(current\_step, decay\_steps)
|
||
|
||
如果 `update_decay_steps` 为true,则每 `decay_steps` 更新 `tmp_decay_step` 的值。公式为:
|
||
|
||
.. math::
|
||
tmp\_decay\_steps = decay\_steps * ceil(current\_step / decay\_steps)
|
||
|
||
**参数:**
|
||
|
||
- **learning_rate** (float) - 学习率的初始值。
|
||
- **end_learning_rate** (float) - 学习率的最终值。
|
||
- **decay_steps** (int) - 用于计算衰减学习率的值。
|
||
- **power** (float) - 用于计算衰减学习率的值。该参数必须大于0。
|
||
- **update_decay_steps** (bool) - 如果为True,则学习率每 `decay_steps` 次衰减一次。默认值:False。
|
||
|
||
**输入:**
|
||
|
||
- **global_step** (Tensor):当前step数。
|
||
|
||
**输出:**
|
||
|
||
Tensor。当前step的学习率值, shape为 :math:`()`。
|
||
|
||
**异常:**
|
||
|
||
- **TypeError** - `learning_rate`, `end_learning_rate` 或 `power` 不是float。
|
||
- **TypeError** - `decay_steps` 不是int或 `update_decay_steps` 不是bool。
|
||
- **ValueError** - `end_learning_rate` 小于0或 `decay_steps` 小于1。
|
||
- **ValueError** - `learning_rate` 或 `power` 小于或等于0。
|
||
|
||
**支持平台:**
|
||
|
||
``Ascend`` ``GPU``
|
||
|
||
**样例:**
|
||
|
||
>>> import mindspore
|
||
>>> from mindspore import Tensor, nn
|
||
>>>
|
||
>>> learning_rate = 0.1
|
||
>>> end_learning_rate = 0.01
|
||
>>> decay_steps = 4
|
||
>>> power = 0.5
|
||
>>> global_step = Tensor(2, mindspore.int32)
|
||
>>> polynomial_decay_lr = nn.PolynomialDecayLR(learning_rate, end_learning_rate, decay_steps, power)
|
||
>>> result = polynomial_decay_lr(global_step)
|
||
>>> print(result)
|
||
0.07363961
|