Behavioral Cloning Loss

Problem

Compute the behavioral cloning loss as the negative mean log-probability of the demonstrated actions. An empty batch returns 0.0.

Hint

Minimizing this loss is equivalent to maximizing the likelihood of the demonstrations.

case 1

in: [[-0.5,-1]]

expected: 0.75

case 2

in: [[0]]

expected: 0

case 3

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

expected: 1.1666666666666667

case 4

in: [[]]

expected: 0

Starter codePython

def behavioral_cloning_loss(log_probs):
    # Your code here
    pass

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