Gradient Centralization

Problem

Apply gradient centralization to a gradient vector.

Compute the mean of all components and subtract it from each component, so the returned vector sums to zero. An empty input returns [].

Hint

The result has zero mean by construction.

case 1

in: [[1,2,3]]

expected: [-1,0,1]

case 2

in: [[-1,0,1,2]]

expected: [-1.5,-0.5,0.5,1.5]

case 3

in: [[]]

expected: []

case 4

in: [[4,4,4]]

expected: [0,0,0]

Starter codePython

def gradient_centralization(grads):
    # Your code here
    pass

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