Lab 3 — Gradient boosting in production

Goal. Train XGBoost, LightGBM, and CatBoost on the same dataset. Tune hyperparameters. Compare training time, inference time, accuracy. Audit feature importance with SHAP.

What you ship. Notebook with 3 trained gradient-boosting models, performance comparison, and SHAP plots for the best one.

Setup

Install the dependencies (one-time).

# !pip install xgboost lightgbm catboost shap scikit-learn pandas matplotlib numpy

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import xgboost as xgb
import lightgbm as lgb
from catboost import CatBoostClassifier
import shap
from sklearn.model_selection import train_test_split
from sklearn.metrics import roc_auc_score
from sklearn.datasets import fetch_openml
import time

np.random.seed(42)

A real-but-public tabular dataset (substitute the Cameroon bank-loan when available)

data = fetch_openml('credit-g', version=1, as_frame=True)
X = data.data
y = (data.target == 'good').astype(int)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25, random_state=42)
print('train/test:', X_train.shape, X_test.shape)

Exercise 1 — Train all three boosters

# YOUR TURN
# Train XGBoost, LightGBM, CatBoost with default hyperparameters.
# Time each fit. Report test AUC.

Exercise 2 — Tune the best one

# YOUR TURN
# For the best baseline, do a grid search over n_estimators, learning_rate, max_depth.
# Report tuned AUC and how much the gain was.

Exercise 3 — SHAP interpretation

# YOUR TURN
# Compute SHAP values for the tuned model on the test set.
# Plot the global feature-importance summary. Identify one feature whose effect
# is monotonic and one whose effect is non-monotonic.

Done?

Submit per the cohort schedule. Peer review pairing announced the following Monday.