GAIL Discriminator Loss Lite

Problem

Compute the binary cross-entropy loss of a GAIL discriminator.

Expert samples have target 1 and policy samples target 0: L = -mean(log(D(expert)) + log(1 - D(policy))). Probabilities are strictly between 0 and 1 and the batch is non-empty.

Hint

The discriminator tries to distinguish expert transitions from policy transitions.

Test cases

case 1

in: [[0.9,0.8],[0.1,0.2]]

expected: 0.328504066972036

case 2

in: [[0.5],[0.5]]

expected: 1.3862943611198906

case 3

in: [[1,0.99],[0.01,0.01]]

expected: 0.015075503780252176

Starter codePython

import math
def gail_discriminator_loss(d_expert, d_policy):
    # Your code here
    pass

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