Pinball Loss

Problem

Compute the mean pinball (quantile) loss for a forecast at quantile q. For each pair add q * (actual - forecast) when actual >= forecast, otherwise (q - 1) * (actual - forecast).

Return the average over all pairs, or 0.0 for an empty actual list.

Hint

The penalties on the two sides of the forecast are asymmetric unless q = 0.5.

Test cases

case 1

in: [[10,10],[12,8],0.5]

expected: 1

case 2

in: [[5],[7],0.9]

expected: 0.19999999999999996

case 3

in: [[5],[3],0.9]

expected: 1.8

case 4

in: [[1,2,3],[1,2,3],0.75]

expected: 0

Starter codePython

def pinball_loss(actual, forecast, q):
    # Your code here
    pass

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