Thank you so much for sharing this project. My friend recommended me your program as a learning material for OFC and I'm having lots of fun playing against it. I'm especially surprised you somehow managed to optimize play style against players who abandons table after winning half buy-in.
I have few questions if you don't mind answering. I understand that you might want to sell your project as commercial solver so you might not want to answer these but any kind of hints would be highly appreciated.
1. Would you consider pineapple OFC to be solved or close to be solved? What about pineapple deuce to seven? I have gut feeling that it's more sophisticated game than regular pineapple but would it not matter for self playing AI?
2. How do you deal with all the possible branches (draws) from so early on?
I'm a CS student who is considering graduate studies and I'm thinking of using poker AI as a research topic if possible.
I already have some experience with topic. I wrote simple CFRM solver and nash shove solver for holdem and my impression so far is that draw games with multiple drawing / betting phase (with hidden discards) tends to have much more complex game tree and thus much harder to solve than games with shared board and static hole cards.
I guess OFC is closer to games like chess or go as it doesn't have betting phase but still the amount of chance node I have to deal with seems absurd. Could you please elaborate on how you approximated opponent behavior? Did you just derive it by self play and then learned its behavior through tree search? Is there mathematical guarantee that there exist unique placement that is optimally +Ev at every single draw?
My apologies if my terminology is off. I don't know much about deep learning and English is not my first language.
This program seems really interesting and I'd love to be able to try it but when I've downloaded the folder and when I try to run the executable, it doesn't boot? Literally nothing happens when I try to open the file. Any advice?
