Bigram Perplexity

Problem

Given a 2D table of bigram counts, compute the conditional entropy H(Y | X) in bits and return the perplexity 2^H. Rows with zero total are skipped.

Hint

Perplexity is the exponentiated conditional entropy of the next token.

case 1

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

expected: 2

case 2

in: [[[10,0],[0,10]]]

expected: 1

case 3

in: [[[3,1],[0,4]]]

expected: 1.324675564281052

case 4

in: [[[1,2,3],[3,2,1]]]

expected: 2.7494592739972052

Starter codePython

import math
def bigram_perplexity(counts):
    # Your code here
    pass

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