Gradient Boosting Stump

Problem

Fit a depth-1 regression stump to gradients (residuals). For each threshold midway between consecutive unique feature values, split X into left (x <= t) and right (x > t) and predict each side's mean gradient.

Return [threshold, left_mean, right_mean] for the split with the lowest total squared error; if no split exists, return [0.0, mean, mean]. Empty input returns [0.0, 0.0, 0.0].

Hint

A stump predicts the mean gradient in each leaf; the best threshold minimizes the total squared error.

Test cases

case 1

in: [[1,2,3,4],[1,1,-1,-1]]

expected: [2.5,1,-1]

case 2

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

expected: [1.5,3,1.5]

case 3

in: [[1,1,2,3],[1,3,2,4]]

expected: [2.5,2,4]

case 4

in: [[5],[2]]

expected: [0,2,2]

case 5

in: [[],[]]

expected: [0,0,0]

Starter codePython

def gradient_boosting_stump(X, gradients):
    # Returns [threshold, left_mean, right_mean]
    # Your code here
    pass

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