drl-zh

Why we included this project

This course teaches deep reinforcement learning the way most people actually learn it: by building the algorithms yourself instead of just reading about them. You start with the fundamentals of Markov decision processes and tabular RL, then work up to the ideas behind AlphaZero, RLHF, Decision Transformers, and Dreamer-style world models. Each exercise is a notebook with TODO sections you fill in, and matching solution notebooks sit alongside so you can unblock yourself when you get stuck. A Docker workspace bundles the DRL-ZH AI Companion, a VS Code extension that gives Socratic hints based on exactly which notebook and step you are on, along with TensorBoard, checkpointing, and Ray support for longer training runs. This is a structured curriculum rather than a library you slot into your own project, so treat it as a thorough way to build real RL fluency before you move on to production systems.