Posterior Collapse Check

Problem

Return True when the KL divergence of a latent variable's variational posterior from its prior falls below the given threshold, which signals posterior collapse: the latent is ignored by the decoder.

Hint

Collapsed latents carry almost no information about the data.

Test cases

case 1

in: [0.001,0.01]

expected: true

case 2

in: [0.1,0.01]

expected: false

case 3

in: [0,0]

expected: false

case 4

in: [0.009,0.01]

expected: true

Starter codePython

def posterior_collapse_check(kl_value, threshold):
    # Your code here
    pass

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