Open Side Menu Go to the Top
Register
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc"

11-24-2012 , 03:21 PM
I like to think that you could program two exploitive bots and pit them against each other until they both reach GTO, thus revealing the GTO solution
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 03:39 PM
Quote:
Originally Posted by HotSource28
I like to think that you could program two exploitive bots and pit them against each other until they both reach GTO, thus revealing the GTO solution
some universities have poker bot competitions. they don't often play gto. they're simply reacting to each others strategies in their own predefined ways.
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 03:42 PM
Quote:
Originally Posted by astrobeaver
some universities have poker bot competitions. they don't often play gto. they're simply reacting to each others strategies in their own predefined ways.
I guess it's the 'own predefined ways' part that is the problem then. Well it's not really a problem they built them for a competition not to find GTO.
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 03:51 PM
Quote:
Originally Posted by dean_nolan1
I have heard him mention in interviews and posts that NLHE (Heads Up at least) was pretty much solved in from a Game Theory Optimal perspective.
No.
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 03:52 PM
Quote:
Originally Posted by HotSource28
I like to think that you could program two exploitive bots and pit them against each other until they both reach GTO, thus revealing the GTO solution
all human players are doing it by playing the game

computers would figure it out way more faster though
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 04:07 PM
That NLHE bot is a ****ing sunrunner
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 04:09 PM
Quote:
Originally Posted by HotSource28
I like to think that you could program two exploitive bots and pit them against each other until they both reach GTO, thus revealing the GTO solution
This sort of works; in fact you need to slightly modify this approach in order to avoid getting into infinite loops. Google "fictitious play".

Of course, for something like HUNL you need a loooot of time and memory.
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 04:18 PM
11-24-2012 , 04:22 PM
Also, a brief explanation of Game Theory by Bryce Paradis:

http://forumserver.twoplustwo.com/94...holdem-245479/
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 04:52 PM
What are the chances that the opening betting in GTO Hu NL is different from what humans have accomplished? For example a lot of players are minraising, so does that mean GTO is closer to minraising or is there a nonzero % chance that GTO could be 4x'ing the button? Can one definitely state that a certain range of opening bets are NOT possible for GTO?
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 04:54 PM
Quote:
Originally Posted by themuppets

there's still a question as to whether or not a human player's approximation of GTO would out-win the same player's attempt at an exploitative strategy.).
The right way to beat a mediocre player is to play as close to GTO as possible except in those situations where it is clear that you shouldn't. Essentially combinining the two types above.

Even though I am not a head up player simple logic tells me that an excellent human will likely beat a semifish for more than a bot would. For example say that you noticed that your opponent is folding 30% of his buttons. A bot would not adjust for this and call too often preflop. It would also play a bit too loose from that point forward. It would give up EV to the semi fish because its strategy has to at least break even against a player that raises 100% of the time preflop. Which is a looser strategy than the proper one for this situation. Of course the EV it gives back to thne semifish is no more than the original gift it was handed by the extra preflop folding. But an expert human, unlike the bot, would not return nearly as much of that gift
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 05:30 PM
Quote:
Originally Posted by micros
..................................
.................................................. ............

In regards to NLHE bots: is it, in theory, possible to create a bot which uses exploitative strategies based on population tendencies and adjusts to opponents as they play differently from these initially set tendencies?
Sure. In time you will see more work being published in this area, or simply witness such a bot playing.

In LH there was a bot called something like "Sonia" that played only exploitively after a small number of hands;I do not remember, perhaps as few as 20. Sonya beat many pros in heads up, and simulated ring. There used to be a site where the public could play Sonya but I can no longer find it.

Here is a link to a 2plus article about a yearly bot vs bot competition in LH: Vol 8. No. 8 By Phil Newall

http://www.twoplustwo.com/magazine/i...ompetition.php

Last edited by tuccotrading; 11-24-2012 at 05:48 PM.
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 05:33 PM
Quote:
Originally Posted by Clayton
What are the chances that the opening betting in GTO Hu NL is different from what humans have accomplished? For example a lot of players are minraising, so does that mean GTO is closer to minraising or is there a nonzero % chance that GTO could be 4x'ing the button? Can one definitely state that a certain range of opening bets are NOT possible for GTO?
This is an interesting discussion, mainly because I remember (a few years ago), some of the prominent posters in these forums debating the value of min-raising preflop in HUNL. I wish I could find the thread. Throughout it, a lot of the posters who were opposing the idea, and calling it -EV, were slowly starting to realize that their assessments were wrong, etc...

Those types of discussions are obviously possible moving forward... where we realize how wrong the standard actions are in certain spots. For all we know, humans are a million realizations away from reaching GTO. It'd be interesting to see a GTO bot in play, so we can all sit here and analyze it's plays and see how wrong we were, and see all of the realizations we've missed out on up to this point.
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 06:01 PM
Quote:
Originally Posted by jusgivithere
If this is true, then I reiterate it would be impossible to not be exploitable.

Since the BOT is never adjusting to a range... it can't come up with a proper calling % against a stand pat. Pre draw, If I stand pat, the GTO bot is going to correctly know that my distribution is any 5 cards.... and it is not making any judgement on whether or not i would stand pat without a made hand or whether I would would stand pat with any 5. Using this only info it has, (i stood pat) it can only know my distribution is any 5. It can then find some % that works against any 5, but then I can adjust and not do it with any 5. I realize it is using a strategy that should work against me no matter how often I do it and no matter my holdings when I do the play...but If it doesn't know how often I am standing pat, and doesn't adjust to it, it is exploitable. With me knowing that the GTO bot is calling with some set frequency against a Stand pat, I can figure out that frequency and exploit it. There is no magic amount it can call that will be not exploitable. I can adjust to its frequency but it can not adjust to mine. I realize GTO means that by definition it is not exploitable, which is why i am saying that for this game GTO does not exist.

Basically if the GTO bot calls the exact same % against someone who stands pat every single hand as it would against someone who only does with an actual made hand, and makes no distinction between these polar opposite styles, there will always be a way to exploit it. 5 Card draw is not beatable without adjusting.
Ok, this thread is long and I can't read it all coz it's in english and it will take me hours... So I just to make sure someone will give the following explanation:

The GTO strategy of HU 5cd will be a loose caller against pats. I can guaranty your pat hands (straights, flushes, boats and SF) will make you a nice profit.

WHY?

Because A) GTO will be indifferent to your appeared strategy (i.e. won't change the way it plays no matter what you do) and B) you can play a strategy that stands pat 100% of the time, such as call pre-pat-bet OOP/raise IP (for example). So GTO, in order not to be exploited by such strategy, will call with a surprising % of hands.

The thing is, that once you know GTO's calling vs pat percentages, you will indeed be able to exploit them, but not without GTO EXPLOITING YOU BACK AGAINST THE REST OF YOUR HANDS. It will call you down quite loosely, allowing you to make a profit standing pat, say, 90% of the hands (we already said it have to call enough to at least break even vs 100% pat). but the profit won't be high enough to cover on the loses from the remaining 10%.
To put it more simply- if you'll stand pat a large % of the time for value with your good hands, GTO will crush you because your remaining ranges will be crappy and will fold/lose at showdown way more often than not. If you will add a lot of bluffs you will find out you don't enough fold equity to make it profitable, not even if you will pat all your hands.

And, as mentioned before, it will be insanely hard to find the GTO strategy, but it been proved to exist for every 2 players game.

p.s.
Poker is a complete information game, money wise. Every player is familiar with all the possible strategies and payoffs of his opponents and himself, so he's able to figure his best response in any given situation.
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 06:09 PM
Here is a simple example which illustrates a few important points regarding GTO play.

My Example: Suppose we are playing "Rock, Paper, Scissor".
  • My strategy is to randomize rock, paper, and scissor (this is GTO plan).
  • Your strategy is to randomly select rock and paper.
We play 1,000,000 games and bet $5 on each game. What is my EV?

Spoiler:
Since the GTO strategy is my strategy, you might think that I have an advantage and will show a profit over time. But the reality is that my expected-value is zero. The advantage of my strategy is only that it is not exploitable. GTO does not exploit all other strategies, including non GTO strategies.

A simple simulation in excel can verify this (I did it a few minutes ago).

Next question: Suppose I knew your plan was to randomize rock and paper. How should I play against you?

Spoiler:
If you are randomizing rock and paper, I should play paper every time since I will either win or tie. In the case of 1,000,000 games at $5 each, my expected-value is $2,500,000.

Note that your strategy is not GTO and therefore it is exploitable.

Conclusions:
Spoiler:
  1. GTO is "un-exploitable" but not always the strategy that makes the most money.
  2. GTO does not exploit all other strategies, including non GTO strategies.
  3. In many cases to make money against someone who isn't playing GTO, you must also deviate from GTO, that is, you must play exploitatively.
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 06:32 PM
to the people saying durrr would get crushed, would you guys expect Hoss_TBF (or any top HULHE player) to beat this slumbot?
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 06:36 PM
Quote:
Originally Posted by Do it Right
People confuse GTO with best. GTO doesn't mean 'best'!

The whole point of GTO is to minimize the maximum loss possible against any other strategy. This also has the consequence, in many games where the GTO solution can be easily shown, of minimizing wins. For instance in roshambo, GTO is obviously to randomly select one of rock/paper/scissor - but it is literally impossible to ever show an advantage over any player with the GTO 'solution'.
Grunch. GTO's point is to achieve an equilibrium such that no matter what your opponent does, he can not exploit you by changing strategies. Ergo, you've made your opponent indifferent towards any choice of actions he may possibly take against you. It's not exactly the same as "minimizing the maximum loss possible", since not every game is about minimizing the maximum loss, though in practice it often is.

Last edited by 28renton; 11-24-2012 at 06:42 PM.
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 06:42 PM
Quote:
Originally Posted by TooCuriousso1
to the people saying durrr would get crushed, would you guys expect Hoss_TBF (or any top HULHE player) to beat this slumbot?
no. i mean, i guess its possible since its still imperfect enough that perfect counter strategy is slightly profitable, but it would be pretty hard for a human to be able to come up with and execute that strategy without a ridiculous amount of time/work/information i'd think. i don't really know **** about it though.
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 06:52 PM
Quote:
Originally Posted by Mindy Macready
Here is a simple example which illustrates a few important points regarding GTO play.

My Example: Suppose we are playing "Rock, Paper, Scissor".
  • My strategy is to randomize rock, paper, and scissor (this is GTO plan).
  • Your strategy is to randomly select rock and paper.
We play 1,000,000 games and bet $5 on each game. What is my EV?

Spoiler:
Since the GTO strategy is my strategy, you might think that I have an advantage and will show a profit over time. But the reality is that my expected-value is zero. The advantage of my strategy is only that it is not exploitable. GTO does not exploit all other strategies, including non GTO strategies.

A simple simulation in excel can verify this (I did it a few minutes ago).

Next question: Suppose I knew your plan was to randomize rock and paper. How should I play against you?

Spoiler:
If you are randomizing rock and paper, I should play paper every time since I will either win or tie. In the case of 1,000,000 games at $5 each, my expected-value is $2,500,000.

Note that your strategy is not GTO and therefore it is exploitable.

Conclusions:
Spoiler:
  1. GTO is "un-exploitable" but not always the strategy that makes the most money.
  2. GTO does not exploit all other strategies, including non GTO strategies.
  3. In many cases to make money against someone who isn't playing GTO, you must also deviate from GTO, that is, you must play exploitatively.
If you took the time to consider this example, please consider the following.

First we don't know the GTO solution to poker. But if we did find this GTO solution, why is everyone assuming it would play +EV against any random strategy or even most strategies?!

The example above illustrates that GTO is only guaranted to be un-exploitable, not profitable.

So for example, a GTO bot might only be break-even against someone who goes all-in every time the action comes to him or her! And GTO bot might only be break-even against someone who min-raises every time the action comes to them!

Until we know the GTO strategy we can't be sure it is +EV against most strategies!

Against a true GTO bot Durrr could possibly (probably!?) click it back every time the action comes to his and break-even (excluding rake)!

Important point: Current bots might be hugely +EV against regs, but the current bots are not playing GTO. So slumbot might crush me and everyone in this thread, but that would be due to the AI effectively us.

Last edited by Mindy Macready; 11-24-2012 at 07:02 PM.
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 07:00 PM
Quote:
Originally Posted by Clayton
What are the chances that the opening betting in GTO Hu NL is different from what humans have accomplished? For example a lot of players are minraising, so does that mean GTO is closer to minraising or is there a nonzero % chance that GTO could be 4x'ing the button? Can one definitely state that a certain range of opening bets are NOT possible for GTO?
My guess would be that there are multiple GTO strategies to NL involving different sizings specifically
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 07:14 PM
Quote:
Originally Posted by Steve Haris
My guess would be that there are multiple GTO strategies to NL involving different sizings specifically
This. Also seems likely that some percentage of btn limping is part of a GTO strategy.
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 07:33 PM
If a bot is playing HUNL and trying to play GTO why wouldnt it just push/fold pre ? Correct me if i'm wrong but GTO just means playing unexploitably rather than optimally. If this is true why bother with other lines when u can just jam/fold PF .

Also I think it's possible the amount of randomness, variability and incomplete information in HUNL could potentially make it so the best possible GTO strategy will be too far away what it actually should be to the point where it could be beaten by a few players. As a disclaimer tho I have no **** clue what im talking about
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 07:34 PM
Quote:
Originally Posted by tuccotrading
In LH there was a bot called something like "Sonia" that played only exploitively after a small number of hands;I do not remember, perhaps as few as 20. Sonya beat many pros in heads up, and simulated ring. There used to be a site where the public could play Sonya but I can no longer find it.
I played against Sonia for a bit online. Can't remember where, it was probably around 2 years ago or so. Found the link on this site somewhere. The one thing I noticed is pre-flop Sonia limped on the button a lot instead of always raising like most hu lhe players.
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 07:49 PM
Quote:
Originally Posted by HotSource28
I like to think that you could program two exploitive bots and pit them against each other until they both reach GTO, thus revealing the GTO solution
Think a few of the serious bot makers out there uses this exact strategy actually.

Quote:
Originally Posted by TooCuriousso1
to the people saying durrr would get crushed, would you guys expect Hoss_TBF (or any top HULHE player) to beat this slumbot?
Even the best HU LHE player in the world wouldnt stand a chance imho. 2-3+ years ago sure, but the most recent GTO bots are super strong at HU LHE.
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote
11-24-2012 , 07:52 PM
Quote:
Originally Posted by aaaaaaaa
If a bot is playing HUNL and trying to play GTO why wouldnt it just push/fold pre ? Correct me if i'm wrong but GTO just means playing unexploitably rather than optimally. If this is true why bother with other lines when u can just jam/fold PF .
probably because it would be exploitable at most stack sizes
Hoss_TBF: "All top players use game theory, distributions, bluff ratios etc" Quote

      
m