active library

agentum

A unified framework for sequential decision-making: from classical search to deep RL. All methods are approximations of expectimax with different representation trade-offs.

Started 2026 JavaScript

Resources & Distribution

Package Registries

What Your RL Algorithm Actually Assumes

An interactive blog series exploring how representation choice — not algorithm choice — determines success in sequential decision-making.

Read the series →

The Series

  1. The Infinite Table — Tabular Q-learning and the cost of making zero assumptions
  2. The Features You Choose Are the Assumptions You Make — Linear function approximation and hand-crafted features
  3. The Architecture Is the Prior — Neural networks learn their own features, but the architecture decides what kind
  4. What You Assume vs. What You Compute — The synthesis: model-based vs. model-free, and the AIXI ideal

Each post includes interactive browser demos where you can train agents, toggle features, and watch representations diverge on the same problem.

Run Locally

python3 -m http.server 8000
# Open http://localhost:8000

No build step. No dependencies. Just static HTML, CSS, and vanilla JavaScript.

Run Tests

Open http://localhost:8000/test.html to run all self-tests in the browser.

License

MIT

Discussion