Noisy Net Sample Seeded

Problem

Sample factorized Gaussian noise for a noisy network using rng = random.Random(seed).

Return weights[i] + sigmas[i] * rng.gauss(0, 1) so every parameter gets its own noise scale. The seed makes the sampled parameters reproducible.

Hint

Noisy nets inject learned parametric noise instead of epsilon-greedy exploration.

case 1

in: [[1,2],[0.1,0.2],42]

expected: [0.9855909670422072,1.965419279933696]

case 2

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

expected: [0.9417154046806644]

case 3

in: [[1,1,1],[0.5,0.5,0.5],7]

expected: [0.8720598557761998,1.255715756258257,0.8869519176084476]

Starter codePython

import random
def noisy_net_sample(weights, sigmas, seed):
    # Your code here
    pass

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