drl-wizard 0.1.2

Deep Reinforcement Learning training platform

DRL Wizard

DRL Wizard is a lightweight, modular Deep Reinforcement Learning toolkit for training, comparing, and understanding modern RL algorithms across diverse environments. It provides a clean workflow with a FastAPI backend, Streamlit UI, structured logging, and support for multiple concurrent training jobs.

---

🚀 Features

• Built-in Algorithms
  PPO, TRPO, DQN, Double DQN, Dueling DQN, SAC — with more on the way.

• Environment Support
  Works with Gymnasium environments, Atari (ALE), image-based observations, multi-discrete action spaces, and custom environments.

• Modern Architecture

  • FastAPI backend for job orchestration & real-time streaming
  • Streamlit UI for configuration, dashboards, and experiment comparison
  • SQLAlchemy repository layer with clean separation of concerns
  • Pydantic-based configuration system with auto-generated forms

• Experiment Management

  • Run multiple simulations concurrently
  • Graceful stop/resume handling
  • NDJSON logging (train/eval segments)
  • Manifest tracking and TensorBoard-compatible metrics
  • Downloadable job archives

• Extensible
  Add new algorithms, environments, or visualization components with minimal boilerplate.

---

📦 Installation

pip install drl-wizard

Install with UI:

pip install drl-wizard[ui]

Install with Development:

pip install drl-wizard[dev]

🖥️ Running

• Running the UI & Backend:

drl-wizard-run

• Running Backend:

drl-wizard-api

• Running UI:

drl-wizard-ui