VAE Reparameterization (Seeded)

Problem

Sample a latent vector with the reparameterization trick: z_i = mu_i + exp(0.5 * log_var_i) * eps_i, where each eps_i is drawn from a standard normal. Call random.seed(seed) first and draw eps with random.gauss(0.0, 1.0) in order.

The signature is vae_reparameterize(mu, log_var, seed).

Hint

Sampling eps outside the network keeps the path differentiable with respect to mu and log_var.

Test cases

case 1

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

expected: [0.9417154047,-1.3965781047]

case 2

in: [[1],[0],42]

expected: [0.8559096704]

case 3

in: [[0],[0],7]

expected: [-0.2558802884]

case 4

in: [[2,-1],[1,0],123]

expected: [2.6664600161,-0.8619886074]

Starter codePython

import math
import random
def vae_reparameterize(mu, log_var, seed):
    # Your code here
    pass

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