Generalization Bound

Problem

Return the Rademacher generalization bound:

true risk <= empirical risk + rademacher + sqrt(ln(1 / delta) / (2 * m))

Return just the two-term complexity penalty added to the empirical risk.

Hint

More samples shrink the confidence term as 1 / sqrt(m).

case 1

in: [0.1,100,0.05]

expected: 0.22238734153404083

case 2

in: [0,50,0.1]

expected: 0.15174271293851463

case 3

in: [0.2,10,0.01]

expected: 0.6798525912188081

case 4

in: [0.05,1000,0.2]

expected: 0.07836756873997225

Starter codePython

import math
def generalization_bound(rademacher, m, delta):
    # Your code here
    pass

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