Edreate.com – Deep Reinforcement Learning Course

Welcome 👋
This repository contains the codebase used in lessons from Edreate’s Deep Reinforcement Learning (DRL) course.

👉 For the full learning experience—including in-depth write-ups, mathematical formulas, video explanations, and structured chapters—visit the course page:
🔗 edreate.com/courses/deep-reinforcement-learning

---

🤝 Community

Join our Discord server for learning, collaboration, and Q&A.

---

🚀 Setup Instructions

For complete setup details, see:
Setting Up Coding Environment and Dependencies

Quickstart (TL;DR)

You’ll need Python and uv installed.

# install uv (if not already installed)
pip install uv

# install all dependencies into .venv
uv sync

# activate the virtual environment
source .venv/bin/activate

# launch Jupyter
uv run jupyter notebook

💡 You can also use your favorite code editor (VS Code, PyCharm, etc.).

---

🌟 Algorithms Covered (Course Highlights)

This repository tracks the main algorithms from the Deep RL course.
Completed ones link to full lessons, others are marked Coming Soon!

✅ Available Now

• Tabular Q-Learning – start with the introductory notebook and walk through a simple 2×3 grid world, then try the stochastic/complex variant to stress-test your policy updates (src/q-learning/q_learning.ipynb, q_learning_2x3_simple_world.ipynb).
• Deep Q-Learning Learn how DQN scales beyond Q-tables and train agents directly with neural networks.

🔜 Coming Soon

• Vanilla Policy Gradient (VPG) – direct optimization of stochastic policies
• Actor–Critic (A2C) – combining value functions with policy learning
• Proximal Policy Optimization (PPO) – stable, scalable policy gradients
• Advanced Methods – SAC and more

🚧 More lessons and code will be added as the course grows!

---

🏁 Benchmark & Use Trained Policies

• Benchmark yourself: run the interactive human baseline for Lunar Lander and see how your manual rewards compare.

  uv run python src/human-benchmark/00_human_lunar_lander_benchmark.py

  Use the arrow keys to control thrust and record your scores across episodes.

• Fly trained agents: plug your saved weights into the Lunar Lander viewers in src/run-lunar-lander/.

  • PyTorch: point MODEL_FILE_PATH in LunarLander_in_Action_PyTorch.py to your checkpoint (discrete or continuous) and run:

    uv run python src/run-lunar-lander/LunarLander_in_Action_PyTorch.py

  • ONNX: export your policy and update the ONNX path in LunarLander_in_Action_ONNX.py, then launch:

    uv run python src/run-lunar-lander/LunarLander_in_Action_ONNX.py

---

📄 License

This project is licensed under the terms of the
LICENSE file in the root of this repository.