VAE KL Term (1D)

Problem

Compute the KL divergence from a diagonal Gaussian posterior to the standard normal prior: -0.5 * sum(1 + log_var - mu^2 - exp(log_var)). The signature takes log_var, not sigma.

The signature is vae_kl_term(mu, log_var). Return a scalar.

Hint

The KL is exactly zero when the posterior equals the standard normal.

Test cases

case 1

in: [[0],[0]]

expected: 0

case 2

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

expected: 0.5

case 3

in: [[0],[1]]

expected: 0.3591409142

case 4

in: [[2],[0.5]]

expected: 2.0743606354

Starter codePython

import math
def vae_kl_term(mu, log_var):
    # Your code here
    pass

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