Model Rollout Horizon

Problem

Choose a model rollout horizon from per-step model accuracy.

Reliability after h steps is accuracy^h, so the smallest h with accuracy^h <= tolerance is ceil(ln(tolerance) / ln(accuracy)), floored at 1. Return 0 when accuracy or tolerance is outside (0, 1).

Hint

Compounding model error limits how far imagined rollouts stay trustworthy.

Test cases

case 1

in: [0.9,0.5]

expected: 7

case 2

in: [0.5,0.25]

expected: 2

case 3

in: [0.8,0.8]

expected: 1

case 4

in: [0.99,0.9]

expected: 11

case 5

in: [1,0.5]

expected: 0

Starter codePython

import math
def model_rollout_horizon(accuracy, tolerance):
    # Your code here
    pass

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