Regret Bound Value

Problem

Compute the standard UCB regret bound for a single suboptimal arm with gap delta after T pulls.

Return 8 * log(T) / delta + (1 + pi ** 2 / 3) * delta, using the natural logarithm. Return 0.0 when delta <= 0 or T < 1.

Hint

The logarithmic term dominates as T grows; the linear term accounts for the gap.

case 1

in: [0.1,1000]

expected: 553.0494091319406

case 2

in: [1,10]

expected: 22.71054887764882

case 3

in: [0.5,100]

expected: 75.82765704265769

case 4

in: [0,50]

expected: 0

Starter codePython

def regret_bound_value(delta, T):
    # Your code here
    pass

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