English | 中文

A personal training progress dashboard for KaTrain games. It reads SGF files exported by KaTrain, parses the embedded KataGo analysis, and displays your opening-phase progress through an interactive web UI.

Overview	Game history

✨ Features

• 📊 Per-game statistics: average points lost, std-dev consistency band, best move rate, policy rank
• 📈 Progress charts over time
• 🎯 Move quality histogram (KaTrain's 6-bucket evaluation scale)
• 🔍 Per-game drill-down: interactive Go board, score/winrate trajectory, phase breakdown (opening / midgame / endgame), undo details
• ⚡ Live updates via WebSocket — drop a new SGF into training/ and the dashboard refreshes automatically
• 🌐 Bilingual UI (Chinese / English toggle)
• 💡 Info tooltips on every key metric

🧰 Requirements

• Docker Desktop (Mac, Windows, or Linux)

That's it. No Python environment or Node.js needed on the host.

🚀 Setup

1. Clone the repository

git clone https://github.com/your-username/go-training.git
cd go-training

2. Create the training folder

SGF files are not committed (they are personal). Create the folder the app expects:

mkdir -p training

Copy your KaTrain SGF files into training/. The app will pick them up automatically at startup and watch for new ones while running.

3. Start the dashboard

docker compose up --build

The first build takes about 30–60 seconds while Docker downloads the Python base image and installs dependencies. Subsequent starts are fast.

Open http://localhost:4096 in your browser.

🎮 Everyday use

Once set up, starting and stopping is just:

# Start (in the background)
docker compose up -d

# Stop
docker compose down

To add a new game, save it from KaTrain as usual — the app detects new SGF files in training/ and updates the dashboard within a few seconds without needing a restart.

🗂 Project structure

go-training/
├── training/               # Your SGF files (gitignored)
├── docker-compose.yml      # Service definition, resource limits
├── Dockerfile              # Python 3.12 slim image
├── requirements.txt        # FastAPI, uvicorn, watchfiles, pydantic
└── app/
    ├── main.py             # FastAPI app, REST API + WebSocket endpoints
    ├── models.py           # Pydantic response models
    ├── sgf_parser.py       # Vendored SGF parser (from KaTrain)
    ├── comment_parser.py   # Parses KataGo analysis from SGF comments
    ├── stats_engine.py     # Per-game statistics computation
    ├── sgf_service.py      # Orchestrates parsing pipeline
    ├── game_cache.py       # In-memory cache with mtime invalidation
    ├── file_watcher.py     # Async file watcher + WebSocket push
    └── static/
        ├── index.html      # Vue 3 SPA (no build step)
        ├── js/
        │   ├── app.js      # Vue application + Chart.js rendering
        │   └── i18n.js     # Chinese / English translations
        ├── css/
        │   └── style.css
        └── img/            # Board images from KaTrain (MIT licensed, bundled)

⚙️ Resource usage

The container runs with explicit limits set in docker-compose.yml:

Resource	Limit	Typical usage
Memory	128 MB	~40 MB
CPU	0.5 cores	< 1% at idle

🔌 API

The backend exposes three JSON endpoints if you want to build your own tooling:

Endpoint	Description
GET /api/games	List of all games with summary statistics
GET /api/stats	Aggregate trends across all games
GET /api/game/{filename}	Full detail for one game (trajectories, board moves, undos)
WS /ws	WebSocket — pushes {"type":"update"} when SGF files change

📝 Notes

• Only KaTrain SGF files with embedded KataGo analysis comments are supported. Plain SGF files without analysis will be parsed but show no statistics.
• The dashboard is designed for opening-phase training (games stopped at move 50). Phase breakdown and policy rank metrics are most meaningful in this context.
• SGF files are gitignored — they are personal data.

🤝 Contributing

This project is fully vibe-coded and maintained as a personal tool. Pull requests are not accepted. If you'd like to extend or adapt it, please fork the repository and develop your own version — that's encouraged.

📄 License

MIT © 2026 Yuhuang Hu

This project includes vendored code from KaTrain (MIT License, © 2020 Sander Land) in app/sgf_parser.py.