The smaller you leave the screen (ctrl-), the faster it runs. Full screen is much slower Mutation rate: If descendants are too unpredictable, then decrease this. If you want more change, then increase this. 5 input of -90 distance, -45 degree distance,...90 degree distance.You can change the number of hidden layers in the code by changing the variable l2size in the start block. Output is one number that changes the direction by a bit called steering. I train these one by one because im too dumb to figure out how to do batches and i didn't have enough time. 'steps' means how many steps it can take (every step it moves forward 4 pixel. It slowly goes up to help training Fitness is calculate by 40 invisible checkpoints + survival time x 3 Try to see what speed u can get to. I got to 22 Press 'r' to reset neural network Only weights, no bias bcs im lazy and dumb I need to add more checkpoints as they stop at the the very left so dont expect ur car to learn past that. Jk the algorithm has been improved Why did I make this. Click 'see inside' to see 3 hours of my life wasted away. I want to die This thing sucks anyways
5 hours of work in: Wow, it finally worked. It still sucks tho but it is now v0.2 This was written by me with no help from online. Shout out to the FTC discord ppl who told me I was stupid. 3Blue1Brown was the best teacher of neural networks on the web. I took that video and decided to practice my skills and decided to make this piece of ****************. Scratch wont let me use even mild profanity so i had to substitute it for using assterisks.
