doomformer is a project from alp (@mottopanikeiku) that compiles an integer neural network into a DOOM map. Voodoo dolls do the arithmetic. Light levels and plane heights hold the values. Every value the map exposes is read back and compared against NumPy.

One map puts 1,355 voodoo dolls to work on an MNIST network. The player inks a digit on the floor, presses READ, and the winning digit’s doll crosses a line, lighting the display. The float version of the model scores 97.58 percent.

There is also a museum.wad that runs without Python at all: step on seven floor pads, press READ, and 136 dolls run a small network. Press MUTATE and the level rewires one connection while it runs, so a 7 then reads as a 1.

Another map holds a DOOM marine’s brain, 1,345 dolls that sense the nearest demon and pick an action. It is not playable. One decision takes 22 seconds of game time, and with the demons awake the marine dies in 5.5 seconds without a kill.

See more on GitHub.