Best-of-N Selection

Problem

Score best-of-N candidates. Return [best_index, best_probability], where best_index is the argmax of the scores (ties to the smaller index) and best_probability is the softmax probability of that index under softmax(scores / temperature), computed stably.

The signature is best_of_n(scores, temperature).

Hint

Low temperature concentrates the probability on the top candidate.

Test cases

case 1

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

expected: [2,0.6652409558]

case 2

in: [[1,1],1]

expected: [0,0.5]

case 3

in: [[1,2],0.5]

expected: [1,0.880797078]

case 4

in: [[5,4,3],10]

expected: [0,0.3671654011]

Starter codePython

import math
def best_of_n(scores, temperature):
    # Your code here
    pass

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