Exploration Bonus Schedule

Problem

Cosine-anneal an exploration bonus from beta_start down to beta_min.

Return beta_min + 0.5 * (beta_start - beta_min) * (1 + cos(pi * t / total_steps)). If total_steps is not positive or t is at least total_steps, return beta_min.

Hint

Cosine annealing starts slow, decays quickly in the middle, then eases out.

Test cases

case 1

in: [0,100,1,0]

expected: 1

case 2

in: [50,100,1,0]

expected: 0.5

case 3

in: [100,100,1,0]

expected: 0

case 4

in: [150,100,2,0.5]

expected: 0.5

Starter codePython

import math
def exploration_bonus_schedule(t, total_steps, beta_start, beta_min):
    # Your code here
    pass

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