Threshold Sweep Best F1

Problem

Find the probability threshold maximizing F1. Sweep unique probabilities from high to low, predicting positive when p >= t, and keep the best F1 (ties keep the highest threshold).

Return [threshold, f1]; no positive labels or empty input returns [0.0, 0.0].

Hint

F1 balances precision and recall, so the best threshold need not be 0.5.

Test cases

case 1

in: [[0,0,1,1],[0.1,0.4,0.35,0.8]]

expected: [0.35,0.8]

case 2

in: [[1,1,1],[0.2,0.5,0.8]]

expected: [0.2,1]

case 3

in: [[0,0],[0.1,0.9]]

expected: [0,0]

case 4

in: [[],[]]

expected: [0,0]

Starter codePython

def best_f1_threshold(y_true, probs):
    # Returns [threshold, f1]
    # Your code here
    pass

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