MediumNLPnlp-126

Viterbi vs Posterior Difference

Problem

Compare a Viterbi path with the per-position posterior distributions. posteriors is a list of state-to-probability dictionaries, one per position.

Return the sorted list of positions where the posterior argmax (ties broken alphabetically) differs from the Viterbi path. Positions beyond the path length are ignored.

Hint

Iterate positions and compare the alphabetical argmax with the path entry when it exists.

Test cases

case 1

in: [["A","B","C"],[{"A":0.6,"B":0.4},{"A":0.7,"B":0.3},{"A":0.2,"B":0.8}]]

expected: [1,2]

case 2

in: [["A"],[{"A":0.5,"B":0.5}]]

expected: []

case 3

in: [["A"],[]]

expected: []

case 4

in: [[],[{"A":1}]]

expected: []

Starter codePython

def viterbi_posterior_diff(path, posteriors):
    # Your code here
    pass

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