Deep RL in 60 Days: Curated Lectures, PyTorch, and Pong

What it does

This repository is a structured 60-day learning challenge that pairs curated lecture videos—mostly from DeepMind and UC Berkeley—with original PyTorch implementations of major deep RL algorithms. It walks through Q-learning, DQN variants, policy gradients, A2C, PPO, evolution strategies, and model-based methods, with weekly Jupyter notebooks tested on OpenAI Gym environments like Atari and RoboSchool.

The interesting bit

Rather than dumping code, the author treats the repo as a syllabus: each week assigns specific theory lectures alongside a “project of the week” notebook, so you are expected to read Sutton & Barto and watch David Silver before touching FrozenLake or Pong. It is essentially a self-directed semester compressed into eight weeks.

Key highlights

• Covers the full arc from tabular Q-learning to advanced policy gradients and model-based RL
• Includes working PyTorch implementations of DQN, Double Q-learning, Dueling Networks, Noisy Nets, REINFORCE, A2C, and PPO
• Curated reading lists and must-read papers (including the original DQN and Rainbow papers) for each module
• Targets OpenAI Gym benchmarks: Atari games, RoboSchool, and continuous control tasks
• Also serves as a promotional vehicle for the author’s companion book, Reinforcement Learning Algorithms with Python

Caveats

• The repository is primarily a curated index of external lectures, papers, and books; the original prose is minimal
• A significant portion of the README is dedicated to promoting the author’s commercial book and affiliate-linked course prerequisites
• It is a learning curriculum, not a maintained framework or library, so the notebooks are starting points rather than production tools

Verdict

Good for self-learners who want a rigid weekly schedule and already know basic PyTorch. Skip it if you are looking for a drop-in RL library or a standalone textbook replacement.