Writing about Reinforcement Leraning
Tutorials, deep-dives, and ideas in reinforcement learning. All posts live on Medium. Considering to write more tutorials on Neurotech related technology in the future.
Follow on Medium
@ym1942
Series · Pokémon Red RL
3 parts
3 parts
Training RED with RL to evolve Wartortle — part 3
Introducing a distance-based healing reward and Q-learning with longer look-ahead to complete the full training loop.
03
Training RED with RL to evolve Wartortle — part 2
Adjusting the reward function so RED learns to move between tiles rather than staying frozen in one spot.
02
Training RED with RL to evolve Wartortle — part 1
Building the Pokémon Red gym environment and teaching RED to stay in the grass to encounter wild Pokémon.
01
Series · Gymnasium environments
2 parts
2 parts
Creating a custom Gymnasium environment — part 2
Building a fully custom Pokémon Red environment from scratch using PyBoy and the Gymnasium framework.
02
Creating a custom Gymnasium environment — part 1
How to edit an existing Gymnasium environment and set up the foundation for a custom RL testbed.
01
Standalone · RL concepts
1 post
1 post
A taxonomy of RL algorithms: DQN, actor-critic, PPO
A structured map of model-free RL algorithms and how they relate to each other — useful reference for practitioners.
—
Series · Finite MDPs
3 parts
3 parts
Solving MDPs — part 3: TD learning
Temporal difference learning — learning value functions mid-episode without waiting for a terminal state.
03
Solving MDPs — part 2: Monte Carlo methods
Learning without a model — estimating value functions from sampled episode returns.
02
Solving MDPs — part 1: dynamic programming
The Bellman equation in practice — using iterative policy evaluation and value iteration on FrozenLake.
01