Full boosting loop: assemble a minimal booster
Advanced
The problem, from first principles
One residual tree makes one correction. A booster chains many such corrections, with each new tree seeing what every earlier tree still missed. The fitted model must also replay that same ordered, scaled sequence for unseen inputs.
From theory to code
Implement train_gradient_boosting and predict_gradient_boosting using the existing residual-tree fitter and regression-tree predictor.
Constraints
- Initialize with
float(np.mean(targets))and one copy per training row. - Fit exactly
n_treessequential residual trees; zero trees return an empty list. - Scale every tree contribution by
learning_rateduring training and prediction. - Return
(initial_prediction, trees)and preserve tree order. - Prediction starts from the same scalar for every new input row.
Hints
Open one at a time. Each gives away a little more than the last.
Hint 1
The next tree must observe the updated predictions, not the original constant prediction.
Hint 2
Training and inference differ only in where the initial prediction and trees came from; both add the same scaled tree outputs in order.
Keywords
classical-mlensemblesgradient-boosting
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.