Poker-AI.org

Hi everyone,

I recently came across the CFR+ project from the university of albert, which was used to solve limit holdem. http://poker.cs.ualberta.ca/cfr_plus.html
Looking at the source code, it seems quite well documented. Was wondering if anyone here has tried to make it work for no limit holdem.

Ha I wish. Most of these things are not available publicly if anyone has worked on them. Because either it was for a competition and still has use for future comps or it was made with some commercial motivation and remains hidden away until they can no longer make money from it

If you have a https://en.wikipedia.org/wiki/Sunway_TaihuLight and a few hundred years that idea might work.

What exactly made that method out of reach, what were the changes between Cepheus and Liberatus for it to handle NLHU games?

Abstraction. All NL Bots use a sort of card abstraction (meaning they treat similar hands as if they were absolutelly equal) and they also restrict the number of actions the bot can recognize, so for instance it might recognize 2/3 pot and pot, but noch 5/6 pot.

NL has a lot more possible states than FL, it will probably not be solved in our lifetime.

CFR+ is not really good for abstracted games btw. it's pretty hard to use imperfect recall cards abstraction.