AIC and BIC

Problem

Compute the Akaike and Bayesian information criteria for a linear model: AIC = n*ln(RSS/n) + 2k and BIC = n*ln(RSS/n) + k*ln(n), where k is the number of parameters.

Clip RSS to at least 1e-12 before taking the log. Return [aic, bic]; n <= 0 returns [0.0, 0.0].

Hint

Both criteria reward fit through ln(RSS/n); BIC punishes parameters more heavily via ln(n).

Test cases

case 1

in: [10,2,3]

expected: [-10.094379124341003,-9.186623845358866]

case 2

in: [100,50,2]

expected: [-65.31471805599453,-60.104377684018345]

case 3

in: [5,0,1]

expected: [-144.20229514181324,-144.59285722937915]

case 4

in: [0,1,1]

expected: [0,0]

Starter codePython

import math
def aic_bic(n, rss, k):
    # Returns [aic, bic]
    # Your code here
    pass

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