Mish

Problem

Apply the mish activation element-wise: f(x) = x * tanh(softplus(x)), where softplus is computed stably as max(x, 0) + log1p(exp(-|x|)).

The signature is mish(x).

Hint

Mish is x times the tanh of softplus, so it is smooth and non-monotonic near zero.

case 1

in: [[0]]

expected: [0]

case 2

in: [[1,-1]]

expected: [0.8650983883,-0.3034014614]

case 3

in: [[2,-2,0.5]]

expected: [1.9439589595,-0.2525014827,0.3752452113]

case 4

in: [[100,-100]]

expected: [100,0]

Starter codePython

import math
def mish(x):
    # Your code here
    pass

4 test cases · run your code against them in the editor