Adversarial Domain Loss

Problem

Compute the binary cross-entropy of a domain classifier.

Apply the sigmoid to each logit with probability clamped to [1e-12, 1 - 1e-12], then average -(y * log(p) + (1 - y) * log(1 - p)) over the samples. Round to 4 decimal places.

Test cases

case 1

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

expected: 0.5032

case 2

in: [[0,0],[0,0]]

expected: 0.6931

case 3

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

expected: 0.1269

Starter codePython

def domain_loss(logits, labels):
    # Your code here
    pass

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