Poker-AI.org

After trying my hand at writing a hand evaluator in Java and getting some poor results I've been looking around at what other people have done. From what I can make out the Ray W 2+2 algorithm (from years back) is the best. Is this still the case? If it is does anyone have code that shows how it is used? I have the PerformanceTest2 class (viewtopic.php?f=24&t=2390) bundled with the code but it doesn't really show how the hands are created and evaluated.

Thanks
Leon

Have you read the remainder of the thread you referenced?

Thanks for the reply. I thought I had but now I see it was a treasuretrove.

I've got something up and running now giving several million evaluations a second.

I wanted to benchmark against Pokerstove with hero having AA and villian having 100% range. I think this is 2,598,960 * 1,326 ~3.5 billion combinations.

Pokertove completes this enumeration in a second claiming several billion games per second.

I don't see how I can ever get near this figure, at the moment my program takes several minutes.

When Pokerstove switches to Montecarlo it suddenly drops to millions of games per second.

I can only assume Pokerstove has some clever way of discounting some of those billion combinations in evaluate mode? Do you know?

That is correct, but I don't know how he does it. Andrew Prock did vaguely describe it once but I can't find it now. I have a suspicion that you can't use that approach for all possible problems though. The source is out there.

You should get over 100 million evals / sec

Have anyone tried snapcall?
https://github.com/platatat/SnapCall

https://poker-ai.org/archive/pokerai.or ... &hilit=lut might interest you.