Confidence Margin

Problem

Compute the confidence margin: the largest softmax probability minus the second largest for a vector of logits. Use a stable softmax.

The signature is confidence_margin(logits).

Hint

A large margin means the model strongly prefers one class over the runner-up.

case 1

in: [[1,1]]

expected: 0

case 2

in: [[2,0]]

expected: 0.761594156

case 3

in: [[0,0,0]]

expected: 0

case 4

in: [[10,0]]

expected: 0.9999092043

Starter codePython

import math
def confidence_margin(logits):
    # Your code here
    pass

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