分段激活 Function

Question

I'm trying to write a piecewise activation function whose slope between -6 and 0 is 0.1 and the other places are one. And the input(X) size is (B, C, H, W). So I concluded that the best way is the simple line code:

 x[-6<x and x<0] = x[-6<x and x<0] * 0.1

But I face this error:

RuntimeError: bool value of Tensor with more than one value is ambiguous

Is there any solution for solving this error?

Answer 1

The most simple version of what you need is:

import torch

def custom_activ(input):
    return torch.where((input>-6) & (input<0.) , 0.1*input, input)