mirror of
https://github.com/TheAlgorithms/Python.git
synced 2024-11-27 23:11:09 +00:00
38 lines
966 B
Python
38 lines
966 B
Python
|
"""
|
||
|
|
Softplus Activation Function
|
||
|
|
|
||
|
|
Use Case: The Softplus function is a smooth approximation of the ReLU function.
|
||
|
|
For more detailed information, you can refer to the following link:
|
||
|
|
https://en.wikipedia.org/wiki/Rectifier_(neural_networks)#Softplus
|
||
|
|
"""
|
||
|
|
|
||
|
|
import numpy as np
|
||
|
|
|
||
|
|
|
||
|
|
def softplus(vector: np.ndarray) -> np.ndarray:
|
||
|
|
"""
|
||
|
|
Implements the Softplus activation function.
|
||
|
|
|
||
|
|
Parameters:
|
||
|
|
vector (np.ndarray): The input array for the Softplus activation.
|
||
|
|
|
||
|
|
Returns:
|
||
|
|
np.ndarray: The input array after applying the Softplus activation.
|
||
|
|
|
||
|
|
Formula: f(x) = ln(1 + e^x)
|
||
|
|
|
||
|
|
Examples:
|
||
|
|
>>> softplus(np.array([2.3, 0.6, -2, -3.8]))
|
||
|
|
array([2.39554546, 1.03748795, 0.12692801, 0.02212422])
|
||
|
|
|
||
|
|
>>> softplus(np.array([-9.2, -0.3, 0.45, -4.56]))
|
||
|
|
array([1.01034298e-04, 5.54355244e-01, 9.43248946e-01, 1.04077103e-02])
|
||
|
|
"""
|
||
|
|
return np.log(1 + np.exp(vector))
|
||
|
|
|
||
|
|
|
||
|
|
if __name__ == "__main__":
|
||
|
|
import doctest
|
||
|
|
|
||
|
|
doctest.testmod()
|