The long-awaited Goise Neural Network. Untrained. Incredibly slow. But it learns. https://turbowarp.org/1195343836?fps=250&stuck&turbo Heavily reduced from the original version. Originally had 2 internal layers of 447 inputs and 50 species, has been reduced to 10 species with 1 internal layer of 225 inputs. Fun. If it doesn't make progress for 2.5 seconds it ends early and moves on to the next species Code is a mess cuz I keep fixing bugs in horridly inoptimal ways I think between this and my influence in adding some Speedrun categories I've definitely cemented my place is GOISE history
Press Space to show/hide variables KlimbRio1 for https://scratch.mit.edu/projects/632428770/ as the savestates helped me figure out what inputs were important and loose inspiration for the thumbnail text Blastmagic for https://scratch.mit.edu/projects/954805734/ to make it run marginally faster/ helping me removing extraneous inputs, for https://scratch.mit.edu/projects/1000421062/ as I used that run to create a set of nodes to score the AI with, and for volunteering to use his computer to run this for a while The way it's scored is it's farthest, plus how far it was at the end, divided by 2