PBT Explore-Exploit Decisions

Problem

Decide the population-based training action for every member.

Return -1 when the score is below bottom_q (copy from a better member), 1 when it is above top_q (keep and perturb), and 0 otherwise.

Hint

The quantile thresholds are strict: scores exactly on a boundary are kept.

case 1

in: [[1,5,10],2,8]

expected: [-1,0,1]

case 2

in: [[5,5],2,8]

expected: [0,0]

case 3

in: [[],1,2]

expected: []

Starter codePython

def pbt_decisions(scores, bottom_q, top_q):
    # Your code here
    pass

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