MediumNLPnlp-156

QA Start/End Span Scores

Problem

Rank all valid QA spans. For each start i and end j with j >= i, the span score is start_logits[i] + end_logits[j].

Return [i, j, score] triples sorted by descending score, then ascending i, then ascending j.

Hint

Enumerate every i <= j pair, then sort with the composite key.

case 1

in: [[0.5,0.1],[0.4,0.2]]

expected: [[0,0,0.9],[0,1,0.7],[1,1,0.30000000000000004]]

case 2

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

expected: [[0,1,2],[0,0,1],[1,1,1]]

case 3

in: [[],[]]

expected: []

case 4

in: [[0.5],[]]

expected: []

Starter codePython

def qa_span_scores(start_logits, end_logits):
    # Your code here
    pass

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