Mutual Information Gap

Problem

Return the mutual information gap (MIG) score for one factor:

MIG = (largest MI - second largest MI) / H(v)

where the MI values are the estimated mutual information between the factor and each latent dimension. When only one value is given the second largest is 0.

Hint

A high MIG means the factor is captured by only one latent dimension.

Test cases

case 1

in: [[0.5,0.2],1]

expected: 0.3

case 2

in: [[0.3,0.3],1]

expected: 0

case 3

in: [[2,1,0.5],2]

expected: 0.5

case 4

in: [[0.1],1]

expected: 0.1

Starter codePython

def mutual_information_gap(mi_values, entropy_v):
    # Your code here
    pass

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