Conformal Prediction Interval Lite

Problem

Build a conformal prediction interval around a point prediction. Take the alpha-quantile of absolute calibration scores (sorted index ceil((1 - alpha) * n) - 1, clamped to [0, n-1]) and return [prediction - q, prediction + q].

Return [prediction, prediction] when scores is empty.

Hint

Conformal intervals are symmetric here because absolute scores are used.

Test cases

case 1

in: [5,[1,-2,3],0.5]

expected: [3,7]

case 2

in: [0,[0.1,0.2,0.3,0.4],0.75]

expected: [-0.1,0.1]

case 3

in: [10,[],0.5]

expected: [10,10]

case 4

in: [2,[1,1],1]

expected: [1,3]

Starter codePython

import math
def conformal_interval_lite(prediction, scores, alpha):
    # Your code here
    pass

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