Stochastic Complexity

Problem

Return the stochastic complexity of a model in nats: the negative log-likelihood plus half a log term per parameter:

SC = nll_nats + 0.5 * n_params * ln(n_samples)

Assume n_samples >= 1.

Hint

This is the BIC-style penalty expressed in nats instead of bits.

case 1

in: [10,2,100]

expected: 14.605170185988092

case 2

in: [0,1,1]

expected: 0

case 3

in: [5,3,8]

expected: 8.119162312519753

case 4

in: [2.5,0,50]

expected: 2.5

Starter codePython

import math
def stochastic_complexity(nll_nats, n_params, n_samples):
    # Your code here
    pass

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