mindspore.nn.PReLU

class mindspore.nn.PReLU(channel=1, w=0.25)[source]

PReLU activation function.

Applies the PReLU function element-wise.

PReLU is defined as:

\[prelu(x_i)= \max(0, x_i) + w * \min(0, x_i),\]

where \(x_i\) is an element of an channel of the input.

Here \(w\) is a learnable parameter with a default initial value 0.25. Parameter \(w\) has dimensionality of the argument channel. If called without argument channel, a single parameter \(w\) will be shared across all channels.

The picture about PReLU looks like this PReLU.

Parameters

  • channel (int) – The elements number of parameter. It could be an int, and the value is 1 or the channels number of input tensor x. Default: 1.

  • w (Union[float, list, Tensor]) – The initial value of parameter. It could be a float, a float list or a tensor has the same dtype as the input tensor x. Default: 0.25.

Inputs:

  • x (Tensor) - The input of PReLU with data type of float16 or float32. The shape is \((N,*)\) where \(*\) means, any number of additional dimensions.

Outputs:

Tensor, with the same dtype and shape as the x.

Raises

  • TypeError – If channel is not an int.

  • TypeError – If w is not one of a float, a float list, a float Tensor.

  • TypeError – If dtype of x is neither float16 nor float32.

  • ValueError – If the x is a 0-D or 1-D Tensor on Ascend.

  • ValueError – If channel is less than 1.

Supported Platforms:

Ascend GPU

Examples

>>> x = Tensor(np.array([[[[0.1, 0.6], [0.9, 0.9]]]]), mindspore.float32)
>>> prelu = nn.PReLU()
>>> output = prelu(x)
>>> print(output)
[[[[0.1 0.6]
   [0.9 0.9]]]]