Added Squareplus Activation Function (#9977)

* Added Squareplus Activation Function

* Added parameter beta to the function

* Fixed Squareplus Function

* Update neural_network/activation_functions/squareplus.py

---------

Co-authored-by: Tianyi Zheng <tianyizheng02@gmail.com>
This commit is contained in:
Kausthub Kannan 2023-10-08 21:38:37 +05:30 committed by GitHub
parent 81b29066d2
commit a12b07f352
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

View File

@ -0,0 +1,38 @@
"""
Squareplus Activation Function
Use Case: Squareplus designed to enhance positive values and suppress negative values.
For more detailed information, you can refer to the following link:
https://en.wikipedia.org/wiki/Rectifier_(neural_networks)#Squareplus
"""
import numpy as np
def squareplus(vector: np.ndarray, beta: float) -> np.ndarray:
"""
Implements the SquarePlus activation function.
Parameters:
vector (np.ndarray): The input array for the SquarePlus activation.
beta (float): size of the curved region
Returns:
np.ndarray: The input array after applying the SquarePlus activation.
Formula: f(x) = ( x + sqrt(x^2 + b) ) / 2
Examples:
>>> squareplus(np.array([2.3, 0.6, -2, -3.8]), beta=2)
array([2.5 , 1.06811457, 0.22474487, 0.12731349])
>>> squareplus(np.array([-9.2, -0.3, 0.45, -4.56]), beta=3)
array([0.0808119 , 0.72891979, 1.11977651, 0.15893419])
"""
return (vector + np.sqrt(vector**2 + beta)) / 2
if __name__ == "__main__":
import doctest
doctest.testmod()