Focal Loss (Binary)

Problem

Compute the binary focal loss for a single prediction.

With p_t = prob when target is 1 and 1 - prob otherwise, return -((1 - p_t)^gamma) * log(p_t) rounded to 4 decimal places. Clamp p_t to at least 1e-12 before taking the logarithm.

Hint

The modulating factor (1 - p_t)^gamma down-weights easy examples.

Test cases

case 1

in: [0.9,1,2]

expected: 0.0011

case 2

in: [0.1,1,0]

expected: 2.3026

case 3

in: [0.5,0,2]

expected: 0.1733

case 4

in: [1,1,2]

expected: 0

Starter codePython

def focal_loss(prob, target, gamma):
    # Your code here
    pass

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