Brain Go Brrrrrr

I found a bunch of flaws like: not being able to see walls or floors, self hits counting towards your fitness, 5 ray casts that I assumed were higher the closer someone was, was actually lower the closer they were, hitting someone had no time limit on how long after that they have to get knocked off for it to count as a knock out for you, if you aim just right you can make your pellets despawn really fast which allowed you to shoot waaaay faster then reasonable, brain had too many hidden layers so I switched to 6-16-6 which seems to work better, it was hard to tell the difference between a pellet and the other people(pellet now looks like a shiny turd), evolution was triggered when someone died which would happen to sporadically and often for effective learning(now happens every 30 seconds), and platform was too small making knocking people off take no real intentional effort, the biases were set waaaay too high initially, spamming pellets was a common strategy, to name a few.

Now the AIs are more chill, can tell if and how away the ground is, can see walls, and shooting pellets lowered your fitness somewhat. They seem to develop more reasonable behavior patterns like not shooting unless there is someone in front of you. I will probably fine tune this a bit more before moving on to adding a major and powerful feature of NEAT, which is the ability for the neural network to create and destroy neurons to try and improve fitness... hopefully I don't break it.