lifrordi/DeepStack-Leduc

Implementing Modicum?

Closed this issue · 8 comments

Have you guys read the latest modicum bot from CMU?

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

snarb commented

Who said that "Libratus beet Deepstack by a large margin"?

libratus authors

snarb commented

Can you pls provide the link? I remember that authors of libraries sad that Libratus won over some other bot with significant margin and DeepStack not. So Libratus should be stronger, but they didn't play.

I am pretty sure Libratus authors never said that "Libratus beet Deepstack by a large margin" simply because that's not true. Libratus never played DeepStack.

I am pretty sure Libratus authors never said that "Libratus beet Deepstack by a large margin" simply because that's not true. Libratus never played DeepStack.

Noam has said DeepStack was not good as Libratus, in Video 50:57 [https://www.microsoft.com/en-us/research/video/ai-for-imperfect-information-games-beating-top-humans-in-no-limit-poker/] or slides page 106 https://www.microsoft.com/en-us/research/uploads/prod/2018/10/Beating-Top-Humans-in-No-Limit-Poker-SLIDES.pdf

I am pretty sure Libratus authors never said that "Libratus beet Deepstack by a large margin" simply because that's not true. Libratus never played DeepStack.

Noam has said DeepStack was not good as Libratus, in Video 50:57 https://www.microsoft.com/en-us/research/video/ai-for-imperfect-information-games-beating-top-humans-in-no-limit-poker or slides page 106 https://www.microsoft.com/en-us/research/uploads/prod/2018/10/Beating-Top-Humans-in-No-Limit-Poker-SLIDES.pdf

But he is just talking about head to head performance against the old techniques, what one cares about is exploitability.
Few lines of code can beat these old techniques by 100x more than Libratus or DeepStack (https://arxiv.org/pdf/1612.07547.pdf). Does it mean that these few lines of code are better than DeepStack/Libratus? No.

If you really want to know what is better, simply read both of the papers and pick the technique that you believe works best for you.
At the end of the day, the approaches are very similar - both relying on search in imperfect information games.

well feel free to provide evidence that counters his claim