FIBA Timeout Optimization Pipeline

This repo reproduces the end‑to‑end analysis delivered to Canada Basketball.

pip install -r requirements.txt
python timeout_pipeline.py --pbp "FIBA Database - pbp.csv"

Key steps

1. ETL – load and clean the play‑by‑play CSV.
2. Win‑Probability Model – logistic on score_diff, time_remaining, period_num.
3. Δ‑WPA Regressor – gradient‑boosting with context (run size, timeouts left, etc.).
4. Monte‑Carlo Traffic Lights – bootstrapped residuals → green / yellow / red.
5. Figures – heat‑map, situational bar chart, cumulative gain.

Outputs land in outputs/.