RL Lessons
I'm learning reinforcement learning by building an AlphaZero-style agent for Star Realms, working through a six-session curriculum: value functions → TD learning → neural networks by hand → Monte-Carlo Tree Search → AlphaZero on a toy game → AlphaZero on Star Realms. These lessons are the write-ups.
- Lesson 1 — From bandits to value functions (lesson & assignment)
- Lesson 1 — assignment feedback & solution
- Lesson 2 — TD learning & self-play (lesson & assignment)
- Lesson 2 — assignment feedback & solution
- Lesson 3 — Neural networks by hand (lesson & assignment — in progress)
- Lesson 3 (bonus) — a two-layer net by hand
- Lesson 4 — From evaluation to control (Q-learning & ε-greedy)
- Lesson 5 — UCB1: directed exploration
- Lesson 6 — Monte-Carlo Tree Search (coming soon)
More lessons will appear here as the sessions happen.