Is Omaha hold'em less mathematically solved? - Page 2

On May 09 2012 14:44 NoTalentTom wrote: A poker bot that is able to beat 50nl of limit hold em? Damn time to start programming again.

hey you, long time no see

Solving the game and playing "game theory optimal" isn't even necessarily the best strategy, it's merely the least beatable. A bot playing GTO might win less than an expert in LHE 6max due to not maximizing EV vs a fish, and might not even beat some games due to rake. I don't think FLH will be solved any time soon, but probably will at some point within our lifetimes.

NL and PLO is obviously a lot harder due to stack sizes. It's literally impossible to solve with infinite stack sizes, but even if we apply a reasonable cap, like 10 buyins, it seems really hard. I think NLH would actually be harder to solve than PLO due to more stack possibilities. For example preflop at most you can 3x it up in PLO, but for NL you can 7.41x it or 112.8x it and you would need a percentage of responses for each of these possibilities for each hand, board, and opponent action. So really it depends on how far you want to simplify. I know the game actually is solved for very small stack sizes in push/fold situations, there are some charts for this floating around somewhere.

On May 09 2012 15:25 Defrag wrote: I don't want to be offensive here, but somebody who stays in microlimits for few years with worse and worse results every year claiming the game is 'mathematically solved' is kinda... funny.

On May 09 2012 14:51 dogmeat wrote: dont read stuff posted by retards would be a good start ...

On May 09 2012 20:05 asdf2000 wrote: Show nested quote + On May 09 2012 15:42 Joe wrote: Afaik: 1. Neither of those games is solved or close to being solved. 2. LHE bots are ahead of human players in HU. Not sure about ring bots, they are probably pretty decent too, but not as good as top human players. prove it, because i seriously doubt it

What do you doubt?

On May 09 2012 15:26 VanDerMeyde wrote: There are 4-5 bots winning 2-3bb/100 at Everest fullring games. 5nl to 100nl they play. And they play on shifts (I never been online for the last 1/2 year and one of them was NOT online)

if this is true why hasnt been any action taken from the site to ban them?

prob because they generate rake

http://www.slate.com/articles/health_...e/2011/02/jeopardy_schmeopardy.2.html
Read the second part of the article, and you will have proves for the points Joe put.

One can not say game is solved if he doesn't precise what he means by this. Winning also has to be defined. Do you consider the result of a hu match (like a hu tourney) or a winning rate ?

GTO solved means, that bot can't lose (not considering rake) against every kind of opponent. But this bot may not extract as much from a third player than the top player.

Or does the bot have to have the best winning rate to consider it solved ?

Either way, without asserting whether it is solved or not, we can clearly say that for every case (hu or ring for lhe, nlhe, plo), bots can still get better.

who cares whether the bots can play GTO or not. the problem is that there's bots that's been taking millions off our eco system for years micros to 2000NL. If u look at high stakes limit, its even more pathetic.

Can't find a clear cut definition of what 'game theory optimal' means. Anyone? I assume it means a strategy that is unexploitable (even if you announce to your opponents what it is) that must yield a profit, but it doesn't have to be the most profitable one. Is that it?

On May 10 2012 09:36 tomson wrote: Can't find a clear cut definition of what 'game theory optimal' means. Anyone? I assume it means a strategy that is unexploitable (even if you announce to your opponents what it is) that must yield a profit, but it doesn't have to be the most profitable one. Is that it?

Pretty much, another way to think of it is the strategy where the best counter strategy to it is that actual strategy. It doesn't necessarily have to earn a profit. In a hu game between two supercomputers that both solved poker, each would be playing GTO (which would be the same exact strategy) so each would be losing to the rake.

and yeah I think the phrase "game theory optimal" is something poker players made up.

Nash_equilibrium

On May 10 2012 09:51 NMcNasty wrote: Show nested quote + On May 10 2012 09:36 tomson wrote: Can't find a clear cut definition of what 'game theory optimal' means. Anyone? I assume it means a strategy that is unexploitable (even if you announce to your opponents what it is) that must yield a profit, but it doesn't have to be the most profitable one. Is that it? Pretty much, another way to think of it is the strategy where the best counter strategy to it is that actual strategy. It doesn't necessarily have to earn a profit. In a hu game between two supercomputers that both solved poker, each would be playing GTO (which would be the same exact strategy) so each would be losing to the rake.

I'd rather phrase it as the most unexploitable play. It is not unique. About Nash equilibrium, problem can be asymetric (if I choose strategy A, it forces opponent to choose B in order to minimise loss and then reach the Nash Equilibrium). So it's more like, if Nash Equilibrium is reached, no player can change alone his strategy to maximize his value.

Thinking about it in a HU with rake, it's pretty clear that it will soon turn a symetric coordination game where both bots would just open fold... (or juste raise, and the other one fold), cause it's the only way to reduce the rake...
fun game, here we are.

On May 10 2012 10:20 kingpowa wrote: About Nash equilibrium, problem can be asymetric (if I choose strategy A, it forces opponent to choose B in order to minimise loss and then reach the Nash Equilibrium). So it's more like, if Nash Equilibrium is reached, no player can change alone his strategy to maximize his value.

In poker at least, it can't be asymmetrical because each player has identical options. If there are two different strategies involved at least one player can change his strategy to increase value.

Thinking about it in a HU with rake, it's pretty clear that it will soon turn a symetric coordination game where both bots would just open fold... (or juste raise, and the other one fold), cause it's the only way to reduce the rake...

Open-folding or always folding when an opponent raises is a very exploitable act, and so it isn't an equilibrium. Maybe there would be some way to program bots so they can agree to pacts with each other, but it wouldn't be part of programming to play at equilibrium.

See it as prisoners dilemna. Due to the fact that they both play perfectly against the other, it is symetric. So they are going to win the same, they have same EV. As there is the rake, this EV is negative. so the only way to maximise EV, is to limit the rake, so to not see any flop. This is the most EV for both of them (as in prisoners dilemna, the fact that they both would just say nothing).
And I disagree, when you say it is exploitable. It is actually not a stable equilibrium (exactly as prisoners dilemna). It means that if one starts raising, the other bot has to put volume in pot (either by calling or raising), and as said, it will raise the rake without giving an advantage to one of those two bots.

On May 09 2012 20:24 TalentedTom wrote: Show nested quote + On May 09 2012 14:44 NoTalentTom wrote: A poker bot that is able to beat 50nl of limit hold em? Damn time to start programming again. hey you, long time no see

this needs more love

On May 10 2012 11:03 kingpowa wrote: See it as prisoners dilemna. Due to the fact that they both play perfectly against the other, it is symetric. So they are going to win the same, they have same EV. As there is the rake, this EV is negative. so the only way to maximise EV, is to limit the rake, so to not see any flop. This is the most EV for both of them (as in prisoners dilemna, the fact that they both would just say nothing).

The solution to prisoners dilemma is that each prisoner rats on the other. In any case bots open folding requires a degree of cooperation that is beyond the scope of programming to play GTO.

And I disagree, when you say it is exploitable. It is actually not a stable equilibrium (exactly as prisoners dilemna). It means that if one starts raising, the other bot has to put volume in pot (either by calling or raising), and as said, it will raise the rake without giving an advantage to one of those two bots.

Its true that for each hand each bot has a different strategy, but overall in the game the sum of the strategies is the same. It is possible for both bots to increase their collective EV if they can somehow agree to terms, but the mere possibility of a bot breaking those terms and Grimming his opponent means that the bots can't be playing at equilibrium.

Ok my bad, I wrongly assume that there is in some point cooperation wich is as you said not considered in GTP. So if there is no cooperation I do agree with what you said.