Platt Scaling Parameters

Problem

Return the parameters [A, B] of the logistic calibration p = 1 / (1 + exp(-(A * score + B))) placed so the decision boundary sits at the midpoint of the two class means: A = 1 / (mean_pos - mean_neg) and B = -A * (mean_pos + mean_neg) / 2. Return [0.0, 0.0] when either class is empty or the means are equal.

Test cases

case 1

in: [[0.8,0.9],[0.1,0.2]]

expected: [1.4285714285714284,-0.7142857142857142]

case 2

in: [[1],[1]]

expected: [0,0]

case 3

in: [[],[]]

expected: [0,0]

case 4

in: [[0.5,0.7],[0.3,0.5]]

expected: [5.000000000000001,-2.5000000000000004]

Starter codePython

def platt_parameters(pos_scores, neg_scores):
    # Your code here
    pass

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