mindspore.nn.probability.bijector.PowerTransform

class mindspore.nn.probability.bijector.PowerTransform(power=0.0, name='PowerTransform')[source]

PowerTransform Bijector. This Bijector performs the operation:

\[Y = g(X) = (1 + X * c)^{1 / c}, X >= -1 / c\]

where c >= 0 is the power.

The power transform maps inputs from [-1/c, inf] to [0, inf].

This Bijector is equivalent to the Exp bijector when c=0.

Parameters
Inputs and Outputs of APIs:

The accessible APIs of the PowerTransform bijector are defined in the base class, including:

  • forward

  • inverse

  • forward_log_jacobian

  • inverse_log_jacobian

It should be notice that the inputs to APIs of the PowerTransform bijector should be always a tensor, with a shape that can be broadcasted to that of power. For more details of all APIs, including the inputs and outputs of the PowerTransform bijector, please refer to mindspore.nn.probability.bijector.Bijector, and examples below.

Supported Platforms:

Ascend GPU

Note

The dtype of power must be float.

Raises

  • ValueError – When power is less than 0 or is not known statically.

  • TypeError – When the dtype of power is not float.

Examples

>>> import mindspore
>>> import mindspore.nn as nn
>>> import mindspore.nn.probability.bijector as msb
>>> from mindspore import Tensor
>>> # To initialize a PowerTransform bijector of power 0.5.
>>> powertransform = msb.PowerTransform(0.5)
>>> value = Tensor([1, 2, 3], dtype=mindspore.float32)
>>> ans1 = powertransform.forward(value)
>>> print(ans1.shape)
(3,)
>>> ans2 = powertransform.inverse(value)
>>> print(ans2.shape)
(3,)
>>> ans3 = powertransform.forward_log_jacobian(value)
>>> print(ans3.shape)
(3,)
>>> ans4 = powertransform.inverse_log_jacobian(value)
>>> print(ans4.shape)
(3,)
property power

Return the power index.

Returns

Tensor, the power index.

forward(value)

forward mapping, compute the value after mapping as \(Y = g(X)\).

Parameters

  • value (Tensor) - the value to compute.

Returns

Tensor, the value to compute.

forward_log_jacobian(value)

compute the log value after mapping, namely \(\log(dg(x) / dx)\).

Parameters

  • value (Tensor) - the value to compute.

Returns

Tensor, the log value of forward mapping.

inverse(value)

Inverse mapping, compute the value after inverse mapping as \(X = g(value)\).

Parameters

  • value (Tensor) - the value of output after mapping.

Returns

Tensor, the value of output after mapping.

inverse_log_jacobian(value)

Compute the log value of the inverse mapping, namely \(\log(dg^{-1}(x) / dx)\).

Parameters

  • value (Tensor) - the value of output after mapping.

Returns

Tensor, the log value of the inverse mapping.