mindspore.nn.PolynomialDecayLR

class mindspore.nn.PolynomialDecayLR(learning_rate, end_learning_rate, decay_steps, power, update_decay_steps=False)

Calculates learning rate base on polynomial decay function.

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

$$\begin{array}{rl}decayed\mathrm{\_}learning\mathrm{\_}rate=& (learning\mathrm{\_}rate-end\mathrm{\_}learning\mathrm{\_}rate)\ast \\ & (1-tmp\mathrm{\_}step/tmp\mathrm{\_}decay\mathrm{\_}steps{)}^{power}\\ & +end\mathrm{\_}learning\mathrm{\_}rate\end{array}$$

Where :

$$tmp\mathrm{\_}step=min(current\mathrm{\_}step,decay\mathrm{\_}steps)$$

If update_decay_steps is true, update the value of $tmp\mathrm{\_}decay\mathrm{\_}steps$ every decay_steps. The formula is :

$$tmp\mathrm{\_}decay\mathrm{\_}steps=decay\mathrm{\_}steps\ast ceil(current\mathrm{\_}step/decay\mathrm{\_}steps)$$

Parameters

• learning_rate (float) – The initial value of learning rate.

• end_learning_rate (float) – The end value of learning rate.

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

• power (float) – The power of polynomial. It must be greater than 0.

• update_decay_steps (bool) – If true , learning rate is decayed once every decay_steps time. Default: False .

Inputs:

• global_step (Tensor) - The current step number. Shape is $()$.

Outputs:

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

Raises

• TypeError – If learning_rate, end_learning_rate or power is not a float.

• TypeError – If decay_steps is not an int or update_decay_steps is not a bool.

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

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

Supported Platforms:

Ascend GPU

Examples

>>> 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)
>>> lr = polynomial_decay_lr(global_step)
>>> net = nn.Dense(2, 3)
>>> optim = nn.SGD(net.trainable_params(), learning_rate=lr)