Feature Normalization In FA

Problem

Normalize features with per-feature mean and standard deviation: (x - mean) / (std + epsilon) elementwise. The epsilon prevents division by zero for constant features.

Hint

Normalizing features keeps one large-scale dimension from dominating the gradient.

Test cases

case 1

in: [[2,4],[1,2],[1,2],1e-8]

expected: [0.9999999900000002,0.999999995]

case 2

in: [[0],[5],[0],1e-8]

expected: [-500000000]

case 3

in: [[1,-1],[0,0],[0.5,0.5],0]

expected: [2,-2]

Starter codePython

def feature_normalization(features, mean, std, epsilon):
    # Your code here
    pass

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