Poker-AI.org

I've got a (preliminary) HUNL SNG AI for 25BB tourneys if anybody is interested in testing/using it. It's a static strategy, created with CFRM. The interface is in XML -- you basically just send the server the current game state and get an action in response. As far as I'm concerned, you can keep whatever profits it makes (if any), for now, if you want to test it live. Or, if you have a bot of your own you want to test it against, that's cool too. I'm just looking to gauge it's performance as I continue to develop it.

can you estimate how long it took you to develop it?

Five years.

Is that all? It's still an infant.

_________________
www.bespokebots.com

> 4 years and and still not finished

I don't think the question was for you, but thank you for sharing.

I tried a lot of different ideas, which didn't pan out. I was working on other things as well.

Server is up. Message me for details if you want to join.

The newer, bigger, better strategy is available if anybody wants to try it. It's computed for HUNL SNGs of up to 40bb starting stacks. All stack intervals combined, the strategy is roughly 250gb in size.

It's not yet finished crunching, nor is yet adaptive, but I believe it will still be a viable competitor as it stands. Currently, it beats my previous strategy by about 21bb/100 at the 40bb stack level and will only get better as time goes on. In a few days I'll begin working on various permutations to be used for adaptive/exploitative purposes.

PM me if you're interested.

Hello Nasher, I am very interested to test your strategy with my partypoker bot for Heads Up Hyper Turbos SNG.
5 years ?? What a work !
I would be honored to test the result of such a work !
I would also like to develop an hyper turbo AI and I will be very interested by how to start.
With less than 8 BB, I will use Nash equilibrium.
In order to adapt to each player in a longer run, i think that an interesting variable to keep is the % of fold of a player
which opened 2BB in SB position, which is a typical mistake of players at this format.
Indeed, this is stupid to open 60 chips and fold 70 % of the times facing a shove.
I don't really know how an IA such yours can start, so don't hesitate to explain me some concepts.

PM if interested,

Autofocus

PS : Btw, does anybody know the best way to programmatically join SNG tables ? I don't still have any idea except simulating mouse events.