Fight Foresight

A small MMA outcome-forecasting pipeline: ELO ratings, a handful of career stats, and a PyTorch classifier trained to predict UFC fight outcomes.

See demo/ for the code, data, and charts — start with demo/README.md.

At a glance

• demo/ufc_fights.csv — a single CSV of ~7,800 historical UFC fights (1993–2026), no odds, no picks.
• demo/features.py — ELO ratings + career-stat diffs.
• demo/train.py — PyTorch model, trained on a chronological split.
• demo/visualize.py — ELO trajectories, calibration, and a model-vs-baseline comparison; plain Python, no notebooks, writes PNG/HTML chart files.

Result: ~60% test accuracy / 0.235 Brier from 11 features on one data source, beating a "just pick the higher-ELO fighter" baseline (~56%).