Surprisingly mesmerizing: nitter.vloup.ch/SamBouiss/status…. Cool to see PPO solve a game that's simple enough to anticipate what move it should make at each step.
samxd099 @samxd099
21 May 2019
@OpenAI here’s my snake trained for ~100000 frames, 5 epochs per frame with PPO

May 21, 2019 · 9:33 PM UTC

2
3
40
Replying to @gdb
It was a cool project, I had fun. Thank you guys for putting out the requests for research.