LoL Ban/Pick Recommendation & Win Rate Prediction System

🔧 Summary

This PR includes the full migration of the LoL_BP_Project from local development to GitHub with:

• ✅ Cleaned Git history (previous large-file commits removed)
• ✅ Proper usage of Git LFS for large files:
  • models/pick_lgb_v3.txt
  • data/stats1.csv / data/stats2.csv
  • Other large .csv, .parquet, .pkl files
• ✅ Organized directory structure: data/, models/, scripts/, ui/, output/, reports/
• ✅ Git ignore rules for intermediate files, system cache, and local logs
• ✅ Dev branch is ready to be merged into main

---

📖 Overview

A complete hero recommendation and win rate prediction system for the Ban/Pick phase in League of Legends. Using historical professional match data, the model predicts the top-k recommended champions and the blue side's real-time win rate at each draft step. The system includes a full Streamlit UI that accepts pick/ban inputs and returns live recommendations and win rate curves.

📊 View Project Presentation

📂 Project Structure

LoL_BP_Project/
│
├── data/
│   ├── matches.csv / participants.csv / teamstats.csv / teambans.csv
│   ├── stats1.csv, stats2.csv        # OraclesElixir match data
│   └── champs.csv                    # Champion ID ↔ Name mapping
│
├── models/
│   ├── pick_lgb_v3.txt               # Pick recommendation model
│   ├── pick_encoder_v3.pkl           # Label encoder
│   └── wr_lgb_v4.txt                 # Win rate prediction model
│
├── output/
│   ├── match_dataset.parquet
│   ├── pick_dataset.parquet
│   └── champ_wr.parquet              # Per-champion blue/red win rates
│
├── scripts/
│   ├── data_preprocess_v2.py
│   ├── train_pick_v3_tqdm.py
│   ├── train_wr.py
│   └── search.py
│
├── ui/
│   ├── app.py                        # Streamlit frontend
│   └── ngrok/                        # External sharing tool
│
└── reports/
    ├── train_pick_v3.log
    ├── pick_cv_v3.txt
    └── wr_cv_v4.txt

---

✅ Installation

conda activate lolbp
pip install -U streamlit lightgbm pandas scikit-learn

---

✅ Required Files

File Path	Description
models/wr_lgb_v4.txt	LightGBM win rate model (trained by train_wr.py)
output/champ_wr.parquet	Per-champion blue/red avg win rate
data/champs.csv	Champion ID → English name mapping
ui/app.py	Streamlit frontend for recommendations and win rate curve

---

🚀 Usage

1. Data Preprocessing

cd scripts/
python data_preprocess_v2.py

2. Train Win Rate Model

python train_wr.py

3. Train Pick Recommendation Model

python train_pick_v3_tqdm.py --n_trials 10 --num_boost 400

4. CLI Test

python search.py --team blue --ban 1,2,3 --topk 5

5. Launch Web UI

cd ui/
streamlit run app.py

6. (Optional) Ngrok External Sharing

ngrok http 8501

---

🧠 Features

• ML Model — LightGBM + Optuna auto-tuning, Top-5 recommendation accuracy ≈ 67.3%
• Draft Support — Auto-filters already picked/banned champions
• Interactive UI — Displays Top-5 picks with predicted probability and win rate curve
• External Access — Ngrok support for sharing over the internet

---

🎯 Win Rate Prediction Curve (Example)

Step	Blue Picks	Red Picks	Predicted Blue WR
1	[1]	[]	52.1%
2	[1]	[11]	50.3%
3	[1]	[11, 64]	48.7%
...	...	...	...
10	[1,105,55,238,61]	[11,64,103,157,121]	59.8%

---

📦 Git LFS Tracked Files

⚠️ Install Git LFS before cloning: git lfs install

• models/pick_lgb_v3.txt
• models/pick_encoder_v3.pkl
• data/stats1.csv / data/stats2.csv
• output/pick_dataset.parquet
• output/match_dataset.parquet
• output/champ_wr.parquet