GTO Confusion
Posted by Max Lober
Posted by
Max Lober
posted in
Gen. Poker
GTO Confusion
I have read that players like Ike haxton, sauce, and many other top players in poker have a GTO strategy that they use. Here is my understanding of it, please someone correct me if I am wrong. They have created a strategy for each potential spot, and make balanced plays so that their opponents have a difficult time reading there hand.
That being said, does this mean it doesn't matter who they play really? They just play their constructed ranges with their balanced moves in each spot regardless of how the villain plays? If this is the case wouldn't this lead to much less variance because now they're close to eliminating potential mistakes, so the only room for variance is within the cards. Maybe they go on a bad stretch of bad beats, or they balance their range in the wrong spot.
This leads me to believe that a style like this is the best way to play because over time if you are applying the concepts correctly, the worst case scenario is you break even. Is a strategy like this usable at small stakes? Well let me rephrase that, is it more profitable than playing a TAG or LAG style to exploit opponents????
Now here is my last question, where can I learn a balanced strategy? Any book, RIO video, or article recommendations?
Loading 18 Comments...
Nobody is playing anything remotely resembling The GTO strategy. We have no idea what it looks like. So how can we claim that nobody does? By logic. Knowing exactly which action to take everywhere, and knowing the optimal bet size down to a fraction of a big blind, is not something a human player will ever be capable of.
Even the best human players use tons of heuristics where little actual thinking is done. They identify patterns, associate them with some good default action they have either learned by study or from experience, and then they pick that action.
However, you can gain a lot of useful insight by analysing isolated scenarios. We can often draw strong conclusions about what the GTO strategy can not look like. And we might have a pretty good idea about whether or not our candidate strategy is easily exploitable.
For example, it's trivial to deduce that folding the majority of your hands on the BTN in a HU game must be very, very far from the GTO strategy.
At small stakes where your opponents are making big mistakes, it seems like GTO play would not outperform very good exploitative play. But even at small stakes there are very good players, and against them the most profitable approach would definitely use a lot of GTO concepts. So I think the most profitable way to play is a mix of GTO and exploitative play. Probably the same as it is at high stakes, you try to play balanced vs regs, and exploit the (odd) fish.
Obligatory game theory study material
Expert Heads Up No Limit Hold'em Vol 1 & Vol 2 (Tipton)
Applications of No-Limit Hold'em (Janda)
The Mathematics of Poker (Chen/Ankenman)
In my opinion, the best material for stuyding up on game theory applied to poker. Tipton's Vol 2 is not published yet, but will be in a couple of weeks. He also sells a video pack as accompanying study material for Vol 1.
MoP is the "GTO Bible" but it's a slightly hard study. Tipton's and Janda's books are hands-on applications of game theory to NLHE and you should not have problems following their steps. The books are also useful for PLO players.
What does it mean to solve a game, anyway?
Applying game theory to real poker is not conceptually hard, although it can be tedious at times. The problems we set out to solve are often practically impossible to solve, though, due to the size of the game. Even if we restrict ourselves to sub-games (like only looking at turn/river play for a hand involving only two players).
Formally writing up the procedure for solving NLHE or PLO entirely is trivial (see Tipton Vol 1 for an explanation of what "Solving the game" actually means). Arriving at the full solution is very hard. In fact so hard, that it's practically impossible with today's computing power. We can tell computers how to do it, but we will have to wait for a very long time. Good progress has been made in simpler games (HU LHE) where today's GTO bots are very strong (see next section for how they measure that).
Since solving NLHE or PLO entirely is not possible in practice, we make do with toy game solutions and various and sundry approximations. Applying those well to real poker is the challenge, and requires a hands-on approach and a lot of thinking.
How do we know when we're there?
Fun fact:
We can calculate how far a candidate strategy is from GTO (in terms of EV), even if we don't know what the GTO strategy looks like. Here is an explanation of that algorithm, and you can click your way to the paper where it was published: http://forumserver.twoplustwo.com/showpost.php?p=28066755&postcount=62.
It's routine for measuring the strength of LHE bots, although much more demanding for testing NL/PL strategies (and I'm not sure anyone has done that yet).About measuring how close we are to the GTO strategy, a.k.a Scientific Method
Some poker researchers adhere to the principles of good Science, like the University of Alberta Computer Poker Research Group. Others, like the PokerSnowie group, don't. The PokerSnowie guys should have submitted their product to the testing algorithm (or the Annual Computer Poker Competition) before hyping up their "GTO software", but they did not, and likely will not.
And I'd like to take this opportunity to say:
Whenever someone claims to have a software product capable of doing this and that in a GTO way, ask them how they know. If they know what they are doing, they should have an answer. It doesn't have to be a mathematical proof, but underneath it all should be a method that is clearly described and testable/falsifiable for anyone interested in undertaking that task.
Also known as The Scientific Method. If a hyped method doesn't adhere to the principles of Good Science, it's probably snake oil. Especially when there's money to be made by serving customers a loose interpretation of the Truth.
Here is one example of GTO software that seems to do things right:
http://blog.gtorangebuilder.com/
To soon to say anything about the utility of this software, but it looks open, scientific, and promising.
there is an extract of the 2nd volume of Tipton's book. I don't link here but you can read and download in the D&B WEB
lots of good information and discussion.
It seems like everyone thinks it is either one or the other. I would imagine there are plenty of spots where the "GTO" play and the exploitative line are very similar. Although I am just assuming that.
By definition, GTO cannot be exploitive, and the two cannot be the same. GTO is absolute, and does not rely at all on your opponent's tendencies. GTO does not change, and if exploitive play was similar to it, it would not be exploitive anymore.
The link Johnson posted is great for understanding GTO, I strongly suggest reading it.
I understand they are different strategies. I was just suggesting that some plays that an "exploitative" players makes and a "GTO" player makes may be similar. Maybe not for the same reasoning perhaps, but may overlap. If a GTO bot is defending his BB perfectly and a exploitative player thinks "this guys is opening a ton from the btn i'm going to start defending more" they are both doing the same thing in that instance. I was just assuming that there would be more situations that would be similar than a simple blind defend. The fact that exploitative plays change would suggest to me that at some point they overlap with what the GTO optimal strategy would be.
I just read part of the suggested article and this was in bold. Exploitative play is a required part of game theory optimal. isn't that a argument to you first point hubris?
The GTO strategy exploits the Nemesis maximally. To maximize your profit against anyone, play maximally exploitatively against anyone. The strategy you end up with against the Nemesis (and which breaks even against him) is the GTO strategy. It has the nice property that it doesn't lose against anyone, not the Nemesis nor anybody else.
Against players who make lots of mistakes, the GTO strategy will win a lot. Because there are tons of mistakes in poker that are impossible to recover from on later streets. When you make those, you are screwed, no matter how well you play later. The most trivial example is to openfold too much on the BTN HU. Even with perfect play for the hands you do open, you are doomed to lose against the GTO strategy. And there are many mistakes like that in poker.
The GTO strategy crushes humans. Well, we can't prove that, but what data we have support that claim:
Today's HU LHE GTO bots are very strong (and that has been verified by the aforementioned testing algorithm). I don't think any human stands a chance, although I don't have recent data to back that up. What I know is that University of Alberta's Polaris bot beat some of the best human LHE players (Hoss_TBF was on the team) HU in 2008. And Polaris has gotten much stronger since.
So if you know a good approximation to the GTO strategy, you can always fall back on that when you don't know how to exploit this player maximally. And you can expect to do very well. When you know how to do better, leave the safe harbour of the GTO strategy and do better.
I could not have said it better, or even half as well as you. Thank whoever is responsible for the great ZenFish! ;)
The study material ZenFish posted are essential, but probably too complicated to someone like OP who seems to start discovering GTO.
There really should exist a sticky reference to a basic, simple and concise introduction to GTO.
I was trying to find one off the top of my head (here or elsewhere), but couldn't.
It is true that a GTO strategy would be essentially break-even not taking rake into account, but that doesn't mean every spot is neutral EV right if two people would be playing at nash vs. eachother. It would just mean that they would be playing identical strategies when positions are reversed thus the end result would be break-even. Kind of to clarify the "break-even" aspect of playing at a nash-equillibrium, because the more I try to study, it seems like the break-even aspect of a GTO strategy is often placed in the wrong spots by some author's/thinkers/posters.
If you're talking about GTO vs GTO, you're absolutely correct. When someone says "GTO breaks even", he's often talking about the game value when two GTO players meet, in which case they both break even in the long run. But there will be variance of course, and if the game is raked, they both lose.
But I've also seen lots of posts about playing just one street, or a single decision point, where some seem to think that the GTO player's action for that decision will have him break even for that decision. And as you point out, this is generally not the case.
For example, in the classical AKQ toy game (OOP player checks, IP player bets 1 into 2, OOP player folds or calls), the IP player makes 1/18 when he bets ==> 5.56 bb/100 for that decision.
Alright, so I thought thanks a lot. The more you read the guesses of some people the more confusion sets in, atleast for me.
A side question, is there any way of knowing that say a 6-max table full of players all playing at a nash, without rake, would they all end up with 0 bb/100 winrates, in other words would that strategy make up for the blind losses in UTG-BTN?
For any randomly dealt hand at that 6-max table, the blinds' expectation is negative, while the other players have positive expectation, with BTN winning most of all.
This follows from logic, since there is no reason for a player outside the blinds to get involved unless his expectation is positive. If it isn't, the player does best by folding (EV = 0). Since all players outside the blinds have positive expectation, the players in the blinds must have negative expectation, since the game is zero-sum.
The players in the blinds will of course get involved in lots of hands with negative expectation, as long as their expectation is still better than folding and surrendering their blinds (-0.5 and -1 bb, respectively).
Because the game is also symmetric (all players average the same number of hands from each position), the expectation for any player in the game is to break even. This assumes they actually play the same number of hands from each seat. If they play, say, 10 hands and then the game breaks, some players will play more hands outside the blinds than others. These players will have positive expectation for the session.
More than just theory
Last year there was a 2+2 debacle about seating rules exploits in LHE. Some players found a way to do what I just described, so they could systematically play 10 hands and only pay one rounds of blinds (or more generally, pay only n-1 rounds of blinds for every n table rotations).
It was estimated that the expectation of a 10-hands-then-quit strategy was 4 BB/100 (assuming sound play). The worst exploit was to play only 4 free hands then quit, for an expectation of 10 BB/100.
Anyone who has played LHE will know that 4 BB/100 and 10 BB/100 are insane win rates. And it was achievable by sitting all day and playing ultra-short sessions, quit, then rinse & repeat at another table. Stars' seating rules made it possible.
Seating rules, then, are just another part of the game's strategy. If rules allow a player to play more hands outside the blinds, he can beat a table of GTO opponents by exploiting that rule. To avoid getting exploited, the other players must have a counter-seating-strategy. In a highly edge-seeking environment where you will get stabbed in the back if you insist on playing fair (as in, you refuse to angle-shoot, even if rules allow it), you might be better off not playing at all. Or work to make the rules fairer. Which is what that 2+2 thread was all about.hi guys, What do you think about gto-range builder? Is there an free alternative?
Be the first to add a comment