mindspore.nn.CosineDecayLR

class mindspore.nn.CosineDecayLR(min_lr, max_lr, decay_steps)

Calculates learning rate based on cosine decay function.

For current step, the formula of computing decayed learning rate is:

$$\begin{array}{rl}decayed\mathrm{\_}learning\mathrm{\_}rate=& min\mathrm{\_}lr+0.5\ast (max\mathrm{\_}lr-min\mathrm{\_}lr)\ast \\ & (1+cos(\frac{current\mathrm{\_}step}{decay\mathrm{\_}steps}\pi ))\end{array}$$

Parameters

• min_lr (float) – The minimum value of learning rate.

• max_lr (float) – The maximum value of learning rate.

• decay_steps (int) – Number of steps to decay over.

Inputs:

• global_step (Tensor) - The current step number.

Outputs:

Tensor. The learning rate value for the current step with shape $()$.

Raises

• TypeError – If min_lr or max_lr is not a float.

• TypeError – If decay_steps is not an int.

• ValueError – If min_lr is less than 0 or decay_steps is less than 1.

• ValueError – If max_lr is less than or equal to 0.

Supported Platforms:

Ascend GPU

Examples

>>> import mindspore
>>> from mindspore import Tensor, nn
>>>
>>> min_lr = 0.01
>>> max_lr = 0.1
>>> decay_steps = 4
>>> global_steps = Tensor(2, mindspore.int32)
>>> cosine_decay_lr = nn.CosineDecayLR(min_lr, max_lr, decay_steps)
>>> lr = cosine_decay_lr(global_steps)
>>> net = nn.Dense(2, 3)
>>> optim = nn.SGD(net.trainable_params(), learning_rate=lr)