Poker Bot piece

[VanVeen]

And how does an optimal strategy for HULHE differ from a strategy that has a positive expectation vs. a random opponent? What sort of computation is required to identify exploitable tendencies in a random opponent and implement suitable countermeasures? I'll give you a few hints: they're very different and not much at all.

I play shorthanded 5/10nl+ and I crush the games I play. Every decision I make is predicated on estimated hand ranges based on the sum of my opponent's actions. I have to guess how often the hand range of someone raising 12% of their hands is interacting with a random flop to give them hands of varying strengths ("how often can they call a raise? how often will they call a flop raise and fold to a turn bet? how often is my middle pair good and what's the most profitable line given their flop/turn/river tendencies?"). It involves mathematical intuition - I'm just flat out guessing that check-call, check-raise turn on 37/47 cards and call, check-fold on the other 10 is maximally profitable given my opponent's hand range and action tendencies. A computer can do this much better than me. It can calculate *exactly* how their hand range is interacting with certain flops; how often they'll improve on the turn; how often they bet, check, call, and raise and what their *likely* hand ranges are after they choose one of those actions. They can use more sophisticated and accurate heuristics to compare the expectation of different lines given their opponent's hand range. They can more accurately model their own play and balance their actions such that their hand ranges aren't easily exploited. The math is not difficult. Hell, I can do the math - I just can't do it on the fly. A computer can.

In short, Pirc is correct. We haven't seen a winning bot because people smart enough to make one haven't bothered tackling the problem. That isn't to say the UofA group isn't smart enough - they've chosen to invest most of their time on finding "optimal" strategies rather than winning strategies (strategies that simply exploit players).

[Sniper]

[ QUOTE ]
I agree that the most compelling argument against bot software is that if it was fail-safe...why would they share it with everyone??

[/ QUOTE ]

Because they can, and it generates an additional stream of income. I continue to shake my head when people continually make this argument on these particular forums (which are clear examples of people with knowledge sharing it for free).

[ QUOTE ]
I parused the website for WinHoldem and I see no guarantee of winnings...no statistical graphs or charts showing "X"BB/hr winnings, nothing.

[/ QUOTE ]

No one said the winholdem creator was a great marketer [img]/images/graemlins/wink.gif[/img]

Additionally, winholdem is really just a shell for others to create bot AI.

[ QUOTE ]
Why not create a bot that took every bit of online information worldwide into account and put it in charge of your stock portfolio, to decide which companies would be a good buy/sell/hold, based on company history,news releases, quarterly reports, etc...You probably wouldnt trust a bot like that with all your $$. Its just too situation specific.

[/ QUOTE ]

Sorry to tell you, but the vast majority of trade volume on the major exchanges every day are automatically triggered by computers, doing the type of analysis you suggest. (Bots, automation, collusion, etc all rule the stock markets)

[ QUOTE ]
Which leads me to ask...Who do you feel has the edge, a bot at a table playing (supposed) perfect hold 'em, or a quality 2+2er sitting at the same table who knows hes playing a bot?

[/ QUOTE ]

They both have an edge, as they will both be focusing on taking money off the fish at the table [img]/images/graemlins/wink.gif[/img]

[RedManPlus]

[ QUOTE ]

In short, Pirc is correct.We haven't seen a winning bot because people smart enough to make one haven't bothered tackling the problem. That isn't to say the UofA group isn't smart enough - they've chosen to invest most of their time on finding "optimal" strategies rather than winning strategies (strategies that simply exploit players).

[/ QUOTE ]

Exactly correct. Key point.

All acedemic research is focused on "basic science"...
Finding optimal, general purpose approaches to beating poker...
That can be applied to more important applications.

Focusing on a very specific game...
Like Party $2/$4 or $22 SNGs...
Where you have player profiles...
And you can back test on 10000s of tournies or 100000s hands...
Is far easier and much more doable.

I'm sure it's been done successfully.

rm+

[img]/images/graemlins/cool.gif[/img] [img]/images/graemlins/cool.gif[/img] [img]/images/graemlins/cool.gif[/img]