mindspore.ops.PReLU

class mindspore.ops.PReLU[source]

Parametric Rectified Linear Unit activation function.

PReLU is described in the paper Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification. Defined as follows:

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

where \(x_i\) is an element of a channel of the input, w is the weight of the channel.

Note

Scalar or 1-D input x is not supported on Ascend.

Inputs:

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

  • weight (Tensor) - Weight Tensor. The data type is float16 or float32. The weight can only be a vector, and the length is the same as the number of channels C of the input_x. On GPU devices, when the input is a scalar, the shape is 1.

Outputs:

Tensor, with the same type as x.

For detailed information, please refer to mindspore.nn.PReLU.

Raises

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

  • TypeError – If the x or the weight is not a Tensor.

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

  • ValueError – If the weight is not a 1-D Tensor.

Supported Platforms:

Ascend GPU

Examples

>>> class Net(nn.Cell):
...     def __init__(self):
...         super(Net, self).__init__()
...         self.prelu = ops.PReLU()
...     def construct(self, x, weight):
...         result = self.prelu(x, weight)
...         return result
...
>>> x = Tensor(np.arange(-6, 6).reshape((2, 3, 2)), mindspore.float32)
>>> weight = Tensor(np.array([0.1, 0.6, -0.3]), mindspore.float32)
>>> net = Net()
>>> output = net(x, weight)
>>> print(output)
[[[-0.60 -0.50]
  [-2.40 -1.80]
  [ 0.60  0.30]]
 [[ 0.00  1.00]
  [ 2.00  3.00]
  [ 4.0   5.00]]]