GAN Discriminator Loss

Problem

Compute the binary cross-entropy discriminator loss: -mean(log(D(real))) - mean(log(1 - D(fake))), averaged over all samples together. D outputs probabilities in (0, 1).

The signature is gan_discriminator_loss(real_probs, fake_probs).

Hint

The discriminator wants D(real) near 1 and D(fake) near 0.

Test cases

case 1

in: [[0.9],[0.1]]

expected: 0.1053605157

case 2

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

expected: 0.6931471806

case 3

in: [[0.7],[0.3]]

expected: 0.3566749439

case 4

in: [[1],[0]]

expected: 0

Starter codePython

import math
def gan_discriminator_loss(real_probs, fake_probs):
    # Your code here
    pass

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