AI Tries To Learn How To Park

Поделиться
HTML-код
  • Опубликовано: 21 окт 2024

Комментарии • 2

  • @HarshG28
    @HarshG28 3 года назад +1

    What about getting to 1000 or maybe 10000 or maybe 100000 generation... Is that even possible?

    • @dubiousinsights4008
      @dubiousinsights4008  3 года назад +1

      I think so. The main limitation here was that I have this set up to run in real time, and am recording this via a screen capture with OSB studio, and then editing it down to ever 10-20 generations. 240 generations @ 15 seconds per took 1 hour and I think 500 MB of raw video. I'm not sure at what size my programs will break, but I doubt I can push it to a 50 GB raw video
      My plan is to instead do 1 of 2 things
      - have pygame output the images of just the generations I want to keep as PNG files and then combine them to a video
      or
      - run the full simulation behind the scenes to create the neural nets, save them, and then just replay certain generations in real time for the screen capture
      I'm also not sure where NEAT's limits are, but I haven't hit them yet, so I can definitely go more generations