Requires godot python addon/asset/module (too big for this repo, install it through asset library)

highscore: 99
deep Q reinforcement learning
4 inputs - bird y velocity, y distances to bottom and top pipe, x distance to right edge pipe
2 hidden layers, size 80 and 40
learning on random batches of 64 states every 4 frames, with game running on 20-30 fps, 3x sped up during training