parsing all-in bets vs slumbot

[yegorrr]

Sending action to server: b18800 | TRACE | Message from server: {'old_action': 'b300b900c/kb1500b7200', 'action': 'b300b900c/kb1500b7200b18800b19100', 'client_pos': 1, 'hole_cards': ['9c', '8c'], 'board': ['9s', '8s', '4c']} | CRITICAL | Error parsing action b300b900c/kb1500b7200b18800b19100: Bet too small - remaining=1200, min_bet_size=1200, new_last_bet_size=300

[christian-git-md]

Hey, it is a bug in the slumbot player, when one player makes two calls in one street (only possible preflop limp - raise - call) then the betsize is calculated incorrectly.

However, I found that there are also other bugs, and is also unclear if the provided models are even properly trained (see other issue). Im general I think this is a promising implementation, and I would be willing to contribute, but it would be nice to know if this repo is still beeing worked on and what results have been achieved with this.

[yegorrr]

i trained my models from scratch, achieving much smaller loss on validation sets but yea i also found some bugs and the code is messy/confusing.. i was going to rewrite it in c++ anyway, but python contribution is also welcomed :) i tried to contact lucky72s on poker ai forums where he posted about this project but no response. feel free to contact me on telegram @yayegor if you're going to work on this