search

happypepper / DeepHoldem

206 stars 65 forks source link

thoughts on libratus, modicum? #4

Open whatsdis opened 6 years ago

whatsdis commented 6 years ago

First, excellent job this is very inspirational.

Have you read the latest modicum bot from CMU?

Also, Libratus beet Deepstack by a large margin, is there any interest in implementing Libratus/Modicum?

happypepper commented 6 years ago

Afaik, libratus and deepstack never played against each other. Given equal thinking time and computing power, I believe deepstack would win because of the depth limited lookahead. The depth limited lookahead enables It to solve for more bet sizes than libratus in less time. It's also a more scalable method that allows it to run CFR on larger games like 3+ player poker.

Regarding modicum, I only skimmed over the paper. The results seem pretty good as it beat first place in 2018 ACPC, but I noticed that it relies on a blueprint strategy that needs only 5 GB to store. I think the weak blueprint strategy could affect the quality of the solution but I'm not sure. It's probably possible to modify modicum slightly so that we allow more computing resources to create an even stronger strategy.