datenwissenschaften 1.7.3

Core library for Retro Speedlab: A high-performance RL toolkit for classic games.

datenwissenschaften: Retro Speedlab Core 🚀

datenwissenschaften is the core engine powering Retro Speedlab, a high-performance Reinforcement Learning (RL) toolkit for classic video games. Built on top of stable-baselines3 and stable-retro, it provides the underlying infrastructure for training, monitoring, and recording RL agents.

⚠️ Important Note

This package is intended as the internal library for the Retro Speedlab project. For the full experience—including automated runners, training scripts, and comprehensive documentation—please use the main repository:

👉 https://github.com/datenwissenschaften/retro-speedlab

✨ Features

• 🎮 Command Center: A rich terminal dashboard for real-time training metrics.
• 🏋️ Orchestrated Training: Simplified RL workflows and session management.
• ⚡ GPU acceleration: CUDA tuning for PPO plus batched GPU inference for NEAT, with CPU fallback.
• 📊 Smart Callbacks: Automatic checkpointing and replay recording (.bk2).
• 🛠️ Robust Infrastructure: Streamlined environment and ROM management.

🚀 Installation

pip install datenwissenschaften

Configuration

Copy config.example.yaml to config.yaml and adjust its values. Application settings are read from YAML rather than environment variables. APIs that load configuration also accept an explicit config_path when the file is stored elsewhere. Relative paths in the file are resolved relative to the configuration file.

Set training.num_envs to auto to select parallel environment workers from CPU affinity, population size, and the systemd/cgroup memory limit. An explicit positive integer continues to override automatic selection.

Set ui.enable: true to run the local Vue training dashboard at http://127.0.0.1:18080. It charts episode fitness and step counts, shows termination outcomes, and reports environment, PPO, and NEAT configuration details. The ui mapping accepts enable, host, port, and max_episodes values. Dashboard history is restored from and atomically persisted to models/<game>/<savestate>/history.json (relative to the configured models directory). Set host: 0.0.0.0 to listen on all network interfaces; use the machine's IP address rather than 0.0.0.0 when opening the dashboard from another device.

📜 License

This project is licensed under the GNU General Public License v3.0. See the LICENSE file for details.

---

Developed with ❤️ by datenwissenschaften.