My online poker AI adventures

[Askalas]

For those interested, here's a link to my blog where I've recently posted a short recap on how things went with my online poker AI. Hopefully I'll be posting my latest findings there if there is enough interest.

http://jimmycarlsson.blogspot.com

[ihaveAdream]

It's nice to hear that your bot wasn't successful even though you put in a lot of effort developing it.

[Dazarath]

So you're posting about your failed attempts at cheating? I, for one, would like my 30 seconds back.

[DCWGaming]

so your illegal bot couldnt win, and you think there is some kind of foul play going on?

I sure as [censored] hope so. Thats the best anti-bot software there is! Dont even tell them they're caught, just turn on their doomswitch and watch the programmers cry.

[jukofyork]

[ QUOTE ]
For those interested, here's a link to my blog where I've recently posted a short recap on how things went with my online poker AI. Hopefully I'll be posting my latest findings there if there is enough interest.

[/ QUOTE ]

"Either my AI/software is bad, online players are much harder opponents than some of the best poker AI available today, online casino isn't playing fair, or other online players are cheating (collusion,etc). I guess that statistically, it could be that I'm having bad luck but taken the number of games played I seriously doubt it."

From what I just read, you tested your bot vs Poki and found that it beat Poki-Bots easily? I'm sorry to say that if you observe how Poki plays, you will see why you (or a fairly simple bot) can beat Poki so easily (ie: it even makes HUGE mistakes preflop...).

Also, you restarted your test runs vs Poki which didn't give Poki time to adapt, which makes Poki use a pretty much static model (which sucks) - it takes maybe 5k hands for Poki to adapt, and even then if you use different stochastic strategies periodically, it will never get a decent model.

I doubt very much your bot was cheated; I think most likely you just assumed that beating a table full of Poki's should indicate your bot is good...

Juk [img]/images/graemlins/smile.gif[/img]

[callme]

I didnt read your blog, but what the previous poster said, i looked like you optimized your bot to beat Poki on his weaknesses. When developing systems of any kind you shouldnt fall for the trap to look into specific past behaviour to try to predict future outcomes - often its refered as over optimization and will produce bad results on future events on diffrent environmentals than Pokis are - which most players produce.

So my guess would be your bot is just not capable of handling the many diffrent styles encountered since he is too specific...

Oh - and beating Poki is not really difficult, i played him a lot and did win constantly - he does some really strange things sometimes and you can bluff him quite easily.

[Askalas]

Hi Juk!

Well I think your conclusion is fair and logical but there is a little more to it of course. The bots (Poki etc) can be configured in a number of ways such as the degree of opponent modelling that should be taken into consideration versus only math, etc.

I did play around 5k games to test each configuration. To reset the bots memory was just one of the tests. I also tried to switch poki's strategy in the middle of the testrun to make it harder for mine.

However, as I said, I do not completely reject the idea that my AI simply isn't good enought for real play. But I do think that it should have had much better results against such weak opponents (lowest levels).

[ihaveAdream]

I think it's safe to say that your bot simply sucked.