Class-Weighted Log Loss

Problem

Compute binary cross-entropy with class weights: each sample is weighted by w1 if y = 1 and w0 if y = 0, and the total is divided by the sum of weights.

Probabilities are clipped to [1e-15, 1 - 1e-15]. Empty input or zero total weight returns 0.0.

Hint

Upweighting a class makes errors on that class cost more in the loss.

Test cases

case 1

in: [[1,0],[0.9,0.1],2,1]

expected: 0.10536051565782628

case 2

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

expected: 0.6931471805599453

case 3

in: [[0,0],[0.2,0.8],1,3]

expected: 0.9162907318741551

case 4

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

expected: 0

Starter codePython

import math
def weighted_log_loss(y_true, y_pred, w0, w1):
    # Your code here
    pass

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