Contrastive Loss

Problem

Compute the mean contrastive loss over pairs. For similar pairs (label 1) add d^2; for dissimilar pairs (label 0) add max(0, margin - d)^2.

Empty input returns 0.0.

Hint

Similar pairs are pulled together; dissimilar pairs are pushed at least margin apart.

case 1

in: [[0.5,2],[1,0],1]

expected: 0.125

case 2

in: [[0.2,0.2],[1,0],1]

expected: 0.3400000000000001

case 3

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

expected: 0.25

case 4

in: [[],[],1]

expected: 0

Starter codePython

def contrastive_loss(distances, labels, margin):
    # Your code here
    pass

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