zebaro.dev/lab
Lab
Small experiments, kept away from the portfolio so they never stand between a visitor and the work.
Labyrinth
An agent starts knowing nothing and learns this maze by trial and error. Cells light up as it learns what they are worth; the dashed line is the route it would take right now.
Ready. Press start and watch it learn.
- Episode
- —
- Steps
- —
- Found the exit (last 20)
- —
- Exploration ε
- —
- Learned route
- —
Tabular Q-learning: a much smaller take on the maze problem of Neural Network Labyrinth, with no neural network. It runs in a Web Worker, so the page stays responsive.
Command menu
Every page, project and contact, a keystroke away. On a phone it is the menu.
Build info
What this page really runs on: release, commit, framework, how the route is rendered.
One more thing is hidden on the home page. It has stripes.