Hi all! Sorry for not keeping up with every issue and PR. I wasn't prepared for this much interest in the project so quickly, but I'm super excited you all are here! 😊
Overall the repo is still in pretty rough shape but looking to improve it.
Stoked to share some updates that should make the training much easier and accessible!
New faster training script that updates model mid-game for faster learning and massively decreased memory usage.
Global reward scaling in config (fixes issue where early training becomes degenerate)
Improved emulator performance by skipping rendering on most frames (3x speed boost). Thanks to the pyboy devs for finding this!
Hi all! Sorry for not keeping up with every issue and PR. I wasn't prepared for this much interest in the project so quickly, but I'm super excited you all are here! 😊 Overall the repo is still in pretty rough shape but looking to improve it.
Stoked to share some updates that should make the training much easier and accessible!
use_screen_explore: False)extra_buttons) #82