Top-3 Accuracy

Problem

Compute top-3 accuracy: the fraction of rows whose target index is among the three largest logits. Break ties in favor of the smaller index.

The signature is top3_accuracy(logits, targets). logits is a 2D matrix and targets has one id per row.

Hint

Sort indices by descending logit with index as the deterministic tie-breaker, then keep the first three.

case 1

in: [[[1,2,3],[3,2,1]],[2,0]]

expected: 1

case 2

in: [[[1,2,3,4],[4,3,2,1]],[0,0]]

expected: 0.5

case 3

in: [[[0,0,0]],[2]]

expected: 1

case 4

in: [[[5]],[0]]

expected: 1

Starter codePython

def top3_accuracy(logits, targets):
    # Your code here
    pass

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