FightIQ Codex – Unified, Leak‑Safe UFC Prediction Pipeline

Overview

• Unified best of both FightIQ repos into a single, clean, configurable pipeline.
• Strict leakage prevention, time‑series validation, calibrated probabilities, and betting strategy support.
• Robust “upcoming fights” inference with feature alignment and fail‑safe preprocessing.

Quick Start

• Configure: edit fightiq_codex/config/config.yaml.
• Train: python fightiq_codex/scripts/train_baseline.py
• Predict upcoming: python fightiq_codex/scripts/predict_upcoming.py

Key Features

• Regex‑based leak filters + temporal split validation.
• Feature‑type imputation (rolling=0, others=median) with optional missingness indicators.
• Optional calibrated outputs (Platt or Isotonic) for better ROI decisions.
• Ensemble stacking (OOF) ready for advanced use.
• Kelly Criterion sizing with caps for risk control.

Structure

• config/ – YAML config for paths, splits, models, options.
• src/data/ – loaders, splitters, preprocessing, validation.
• src/evaluation/ – metrics.
• src/models/ – ensemble + calibration.
• src/betting/ – Kelly.
• scripts/ – train/predict entry points.
• tests/ – leakage checks.
• artifacts/ – saved models, imputers, scalers, features (created on train).
• outputs/ – predictions and reports.

Notes

• No external downloads; upcoming predictions expect a CSV per config.
• Unicode/emojis avoided to keep logs portable across terminals.