doomformer
An experiment that turns an integer neural network into a DOOM map and computes inside the game
What does it do?
doomformer compiles an integer neural network into mechanisms in a DOOM map. When a user draws a number on the floor and presses READ, voodoo dolls in the map perform the neural network computation and display the resulting number. Python can read 248 intermediate values and compare them with NumPy results at a tolerance of 0. In a browser-based Colab notebook, users can draw numbers without installing anything. In the museum example, they can enter numbers using floor plates and press MUTATE to see how the result changes when 1 connection is altered.
Key features
- Compile an integer neural network into a DOOM map
- Enter numbers in the map and display neural network results
- Compare 248 intermediate values with NumPy at a tolerance of 0
- Try it in a browser through a Colab notebook without installation
- Use MUTATE to alter 1 connection and observe the change in results