How does GTO win?
The phrasing is a bit imprecise, but you are essentially correct.
A NE strategy for an abstract games with a number of possible sizings can't be exploited by a strategy that uses a subset of these sizings. This follows trivially from the definition of a NE, idk how this is even an argument.
A NE strategy for an abstract games with a number of possible sizings can't be exploited by a strategy that uses a subset of these sizings. This follows trivially from the definition of a NE, idk how this is even an argument.
edit: well actually no. we agree on that.
the argument is that aner0 and i disagree on how useful a nash sim from another toy game is. It started out by us discussing whether a 10 sizing sim can be used to counter an opponent who plays a single size cbet.
Well that and the fact that I believe multi preflop sizing sims are a scam since they don't accurately represent the game tree.
Preflop sims should be single size. one at a time. Because no human mixes between 2,2.2 and 2.3x
I've got a pretty simplistic question. Say I want to study a spot using 3 sizing on every street. Is a better to run a single SIM or a different SIM for each sizing? I assume the former because a solver actually cares about what may happen on future streets.
I'm just catching up with the recent posts now though, might be missing some context.
Really all we disagree on is how bad the 10 sizing sim would do if it faces a range cbet of 25%. (in the very specific node of villain b25% only)
I say it's pretty far off even if the 10 sizing sim includes b25%. It's actually a silly argument because "far off" is relative.
I think it's far enough from reality so i won't bother ever running sims with multiple RFI sizes and will never go over 4 sizings for postflop either. works just fine for me
dude45. you want all 3 sizes at the same time for a postflop Sim. 3 is perfectly reasonable for postfllop if that's what you want to study.
If you only want to use one of these sizings in the end I'd strongly encourage you to resolve with that one sizing only in seperate sim. Things will change slightly and it's well worth the effort to make sure you do this properly
There's no more argument there. We all agree that the 10 sizing strat will do better when compared to 1 sizing strat. That's no surprise.
Really all we disagree on is how bad the 10 sizing sim would do if it faces a range cbet of 25%. (in the very specific node of villain b25% only)
I say it's pretty far off even if the 10 sizing sim includes b25%. It's actually a silly argument because "far off" is relative.
Really all we disagree on is how bad the 10 sizing sim would do if it faces a range cbet of 25%. (in the very specific node of villain b25% only)
I say it's pretty far off even if the 10 sizing sim includes b25%. It's actually a silly argument because "far off" is relative.
Solution A allows 10 cbet sizings, one of them is a 25% pot bet.
Solution B uses a single 25% pot cbet sizing.
Except for that cbet sizing the game trees are identical otherwise. We assume both solutions are perfectly converged for their respective abstract games.
If we play that 25% pot cbet strategy from solution B against solution A:
1) EV of player using A >= EV in equilibrium of abstract game A
2) EV of player using B >= EV in equilibrium of abstract game B
The discussion is then essentially about the degree of that second part: How much better does the single bet solution do when playing against the multi bet solution, compared to the single bet equilibrium.
The answer is: Potentially a lot, especially if you are looking at a sizing that is far from optimal. That's the part where your experiences may diverge: If you are looking at a "sane" single bet sizing then the difference will likely be small. Using that single sizing may already realize close to optimal equity, adding other sizes doesn't change much in terms of overall EVs. If you are looking at a "silly" single bet sizing then the difference is probably huge. Solution A doesn't really need to defend against the "silly" sizing: It's sufficient to make sure that the "silly" sizing performs worse than the "sane" alternatives, but that's an incredibly low bar.
Hate to quote my own post but to be clear talking about from the flop forward.
The discussion is then essentially about the degree of that second part: How much better does the single bet solution do when playing against the multi bet solution, compared to the single bet equilibrium.
The answer is: Potentially a lot, especially if you are looking at a sizing that is far from optimal. That's the part where your experiences may diverge: If you are looking at a "sane" single bet sizing then the difference will likely be small. Using that single sizing may already realize close to optimal equity, adding other sizes doesn't change much in terms of overall EVs. If you are looking at a "silly" single bet sizing then the difference is probably huge. Solution A doesn't really need to defend against the "silly" sizing: It's sufficient to make sure that the "silly" sizing performs worse than the "sane" alternatives, but that's an incredibly low bar.
The answer is: Potentially a lot, especially if you are looking at a sizing that is far from optimal. That's the part where your experiences may diverge: If you are looking at a "sane" single bet sizing then the difference will likely be small. Using that single sizing may already realize close to optimal equity, adding other sizes doesn't change much in terms of overall EVs. If you are looking at a "silly" single bet sizing then the difference is probably huge. Solution A doesn't really need to defend against the "silly" sizing: It's sufficient to make sure that the "silly" sizing performs worse than the "sane" alternatives, but that's an incredibly low bar.
The outputs are not "good enough" or "close".
Results are either wrong or right. They are right if your opponent uses the 10 sizing strategy you put into pio. They are wrong if your opponent uses only 1 or 2 sizings. Even if they are covered in your sim.
Also speaking practically for a moment: A 10 sizing pio sim assumes we split our range 11 different ways, using every size a fraction of the time....
do people really run sims like that and/or attempt to play like this? I feel very strongly about limiting my bet size to something I can execute
make sure these bets are spaced out and you don't forget to allow raises.
that's all you need
I know that's not quite what you were talking about, but it's important to keep this in mind.
Regarding practical solves:
Generally I see quite a few "solves" around that are obviously not even remotely close to converged. Using an excessive number of sizings seems like a really bad use of resources in most cases. Rather limit complexity and do a decent solve on that smaller game than pack your RAM and then not have the resources to actually solve that huge game to any reasonable degree of accuracy. And if you really want to increase complexity, I'd still increase abstraction quality over number of sizings first.
Meh, let's not forget that in practice we are all working with approximations of NEs. For most of the larger solves we don't even have an accurate idea about the solutions' exploitability within the abstract games(!). This is "good enough" / "close enough" all over the place.
I know that's not quite what you were talking about, but it's important to keep this in mind.
Regarding practical solves:
Generally I see quite a few "solves" around that are obviously not even remotely close to converged. Using an excessive number of sizings seems like a really bad use of resources in most cases. Rather limit complexity and do a decent solve on that smaller game than pack your RAM and then not have the resources to actually solve that huge game to any reasonable degree of accuracy. And if you really want to increase complexity, I'd still increase abstraction quality over number of sizings first.
I know that's not quite what you were talking about, but it's important to keep this in mind.
Regarding practical solves:
Generally I see quite a few "solves" around that are obviously not even remotely close to converged. Using an excessive number of sizings seems like a really bad use of resources in most cases. Rather limit complexity and do a decent solve on that smaller game than pack your RAM and then not have the resources to actually solve that huge game to any reasonable degree of accuracy. And if you really want to increase complexity, I'd still increase abstraction quality over number of sizings first.
Or while we are being honest I see programs like GTOX and zenith. They compromise on accuracy but add heaps of sizings all over the place. it's utterly disgusting when this is somehow market as "closer to gto, therefore more sizings = better".
i know i know. free market and all that but I still like to call them out for selling garbage
My god if you are a zenith customer that explains everything. Let me put it blundly. They are wrong about it. Their product is basically a scam.
And you fell for it.
Don't believe me? https://forumserver.twoplustwo.com/2...ghlight=zenith
but anyway keep trying to "cover" every size if that's what you want. It won't give you the right answer (and you know that) but it will give you an answer that's "close enough" (whatever that means).
So enjoy your "close enough" answers that are calculated based on ranges that aren't even real. Maybe watch some finding equilibrium while you are at it.
And you fell for it.
Don't believe me? https://forumserver.twoplustwo.com/2...ghlight=zenith
but anyway keep trying to "cover" every size if that's what you want. It won't give you the right answer (and you know that) but it will give you an answer that's "close enough" (whatever that means).
So enjoy your "close enough" answers that are calculated based on ranges that aren't even real. Maybe watch some finding equilibrium while you are at it.
The point of a 10-sizing strategy being "close enough" compared to a 1-sizing strategy is that it's a refutation of the 1-sizing strategy, even if not an optimal one. If I play the perfect counter to a 10 sizing strategy when you're only using a subset of those sizings I'm going to beat you.
well said. Although I want to point out that even "likely a small ev loss" isn't remotely acceptable when talking GTO.
The outputs are not "good enough" or "close".
Results are either wrong or right. They are right if your opponent uses the 10 sizing strategy you put into pio. They are wrong if your opponent uses only 1 or 2 sizings. Even if they are covered in your sim.
The outputs are not "good enough" or "close".
Results are either wrong or right. They are right if your opponent uses the 10 sizing strategy you put into pio. They are wrong if your opponent uses only 1 or 2 sizings. Even if they are covered in your sim.
100% correct and that's exactly what i was trying to get at. I do honestly get triggered when i see multi sizing preflop sims being praised as somehow superior to 1 sizing preflop sims.
Or while we are being honest I see programs like GTOX and zenith. They compromise on accuracy but add heaps of sizings all over the place. it's utterly disgusting when this is somehow market as "closer to gto, therefore more sizings = better".
i know i know. free market and all that but I still like to call them out for selling garbage
Or while we are being honest I see programs like GTOX and zenith. They compromise on accuracy but add heaps of sizings all over the place. it's utterly disgusting when this is somehow market as "closer to gto, therefore more sizings = better".
i know i know. free market and all that but I still like to call them out for selling garbage
What point are you making by linking that thread? Doesn't really expose anything bad about the quality of Zenith's material.
The point of a 10-sizing strategy being "close enough" compared to a 1-sizing strategy is that it's a refutation of the 1-sizing strategy, even if not an optimal one.
The point of a 10-sizing strategy being "close enough" compared to a 1-sizing strategy is that it's a refutation of the 1-sizing strategy, even if not an optimal one.
As for zenith: Top players laugh at the garbage they put out. I wonder why?
I can tell you why: because you can't simply mash 4 preflop sizings into monker and expect to get a good answer.
garbage in. garbage out mate.
GTO "accepts" EV loss compared to exploitative strategies all the time. You're exploiting when you assume an opponent is using a single size strategy and study the counter to that strategy. By this logic your 1 sizing sim against a 1 sizing opponent is still "wrong" if the opponent is range c-betting when your sim says he should be betting 80% and checking 20%.
My point is that for every set of sizing strategies there's a different GTO solution.
Your job is to run sims making the closest possible estimations of your opponents sizing strategy. So by all means:
If you know your opponent tries to use 10 sizes. Run a sim for their size. If you know your opponent single sizes you need a counter for their strategy.
The best way to get the closest possible results compared to what is actually going on is to study using 2-3 flop sizes and a couple more on later streets.
If you ever find yourself looking at ranges that have sub 3 combos by the river well... you are probably not looking at anything that is applicable to the real game.
As for the exploitative debate: assuming an opponent uses a single size strat preflop and playing accordingly is not exploiting. That's called being realistic.
No player mixes in between 4 preflop sizings. And certainly no player mixes 4 preflop sizings according to zenith simulations. Except for the people who fell for their shady marketing.
I did my reseach and watched a bit of their youtube videos, fast forwarding to Thomas revealing their monker settings.
Only 10 buckets per spot. Accuracy set on very low. Sims run for a veryyyy short amount of time (thus not getting in enough iterations).
A better question would be: How can you be sure they are doing it properly? What incentive do they have to not do this?
Looks like they are getting away with rushing their sims just fine. Even smart people on here believe their stuff is good.
But to be clear about it: to accurately solve a multi sizing preflop strategy would take massive resources. The game tree would literally explode.
TLDR: zenith cuts massive corners they shouldn't cut but who cares. people buy their crap anyway
bro why the **** are you still talking about zenith in this thread, CTRL+F "zenith" 20 total results, 19 of those are you (some quoted) and someone else mentioned it once (as a response to you)
as per my initial post in here I feel like we need a sticky in this subforum for GTO, the exact same incorrect info comes up all the time.
as per my initial post in here I feel like we need a sticky in this subforum for GTO, the exact same incorrect info comes up all the time.
bro why the **** are you still talking about zenith in this thread, CTRL+F "zenith" 20 total results, 19 of those are you (some quoted) and someone else mentioned it once (as a response to you)
as per my initial post in here I feel like we need a sticky in this subforum for GTO, the exact same incorrect info comes up all the time.
as per my initial post in here I feel like we need a sticky in this subforum for GTO, the exact same incorrect info comes up all the time.
A) the multi preflop sizing stuff is actually a thing that top professionals believe/apply in the same way they do. so 4 sizes into the preflop solve. then pretend to play a game against the 4 way range split.
B) show evidence that a preflop multi sizing solve can actually be done accurately with today's software. They use monker and the ram requirements for even fairly simple tree is crazy high. Adding more than 2 sizings pre will blow up the tree size massively.
The only way to cut the tree size back down to then make it solveable would be to reduce bucketing/texture.
It's also a fact that reducing monker buckets/texture settings too much will yield poor results.
And then there is also the quite big concern, that they don't run their sims nearly long enough.
I was using them as an example for shitty preflop simulations that sell you on the multi rfi sizing nonsense dream. They seem to be some kind of authority posters here tend believe and it drives me crazy.
If you'd rather i call out another provider of equally bad sims I'd be happy to do that but afaik they are the only ones who do sell this trash.
is anything i said incorrect? By all means please show me proof that:
A) the multi preflop sizing stuff is actually a thing that top professionals believe/apply in the same way they do. so 4 sizes into the preflop solve. then pretend to play a game against the 4 way range split.
B) show evidence that a preflop multi sizing solve can actually be done accurately with today's software. They use monker and the ram requirements for even fairly simple tree is crazy high. Adding more than 2 sizings pre will blow up the tree size massively.
The only way to cut the tree size back down to then make it solveable would be to reduce bucketing/texture.
It's also a fact that reducing monker buckets/texture settings too much will yield poor results.
And then there is also the quite big concern, that they don't run their sims nearly long enough.
I was using them as an example for shitty preflop simulations that sell you on the multi rfi sizing nonsense dream. They seem to be some kind of authority posters here tend believe and it drives me crazy.
If you'd rather i call out another provider of equally bad sims I'd be happy to do that but afaik they are the only ones who do sell this trash.
A) the multi preflop sizing stuff is actually a thing that top professionals believe/apply in the same way they do. so 4 sizes into the preflop solve. then pretend to play a game against the 4 way range split.
B) show evidence that a preflop multi sizing solve can actually be done accurately with today's software. They use monker and the ram requirements for even fairly simple tree is crazy high. Adding more than 2 sizings pre will blow up the tree size massively.
The only way to cut the tree size back down to then make it solveable would be to reduce bucketing/texture.
It's also a fact that reducing monker buckets/texture settings too much will yield poor results.
And then there is also the quite big concern, that they don't run their sims nearly long enough.
I was using them as an example for shitty preflop simulations that sell you on the multi rfi sizing nonsense dream. They seem to be some kind of authority posters here tend believe and it drives me crazy.
If you'd rather i call out another provider of equally bad sims I'd be happy to do that but afaik they are the only ones who do sell this trash.
You **** on defending like zenith preflop sims because people only play 1 sizing and your defense will underperform. But somehow you miss that if you split sizings yourself, literally everyone and their mum who use single sizing defense charts will underperform against you.
Also, there are ways in which you can vastly reduce ram usage for a 6-max preflop sim without simplifying bucketing, texture or anything really.
Their sims are constantly running and rerunning, so you just get the latest, most converged one at the time. And they are not designed for you to learn them by memory, but to understand what's going on from looking at which sizings the solver chooses on which spots.
I have said that time and time again and it's an undeniable fact.
How do you study postflop? Do you keep pretending your opponent uses a multi sizing strat pre and therefore only has a narrow 30% ish range after raising the button?
Do still not see why this is potentially extremely dangerous to run sims with such wildly inaccurate assumptions?
As for the quoted part. What on earth are you on about? You act like the EV in poker comes from playing the most complex preflop sim possible.
i have more sizings preflop. therefore i will gain more EV than you. That's frankly the dumbest thing I have read on here so far. This isn't how poker works.
By the way in order to correctly multisize pre you would need to solve for 2 sizings pre. That's the way to do it. 2 sizings pre should work out fine with decent accuracy.
Also if the strategies are not even designed to be used at the table what are they even for? Why would you study something you can't even use?
NGL aner0 you sound like you are a massive fanboy. Maybe take a step back and ask yourself if the zenith guys are potentially not telling you the truth. They run a business based on selling stuff no one else sells. All while claiming everyone else's work is bad.
Justluck clearly has a hateboner for Zenith.
It won't do as well compared to a solution targeted at their exact simplified strategy, but it will do better against the range of strategies they could potentially play. This is just the classic GTO vs Explo argument.
More bet sizes create more rugged strategies. For example, if BTN assumes that BB can only raise half-pot, then BTN can value bet wider and more linearly as BB won't be able to apply enough pressure. Or let's say BB knows that BTN only cbets 33% on the flop. BB will be able to call wider pre, knowing that BTN can't apply enough pressure on a lot of flops.
My point is entirely correct. Multi sizing sims can't be applied as a one size fit all solution. Even if your silly charts include all the sizes. This applies on the flop and before the flop. Therefore your multi size 3x answer won't do well against someone playing 3x only.
More bet sizes create more rugged strategies. For example, if BTN assumes that BB can only raise half-pot, then BTN can value bet wider and more linearly as BB won't be able to apply enough pressure. Or let's say BB knows that BTN only cbets 33% on the flop. BB will be able to call wider pre, knowing that BTN can't apply enough pressure on a lot of flops.
My point is entirely correct. Multi sizing sims can't be applied as a one size fit all solution. Even if your silly charts include all the sizes. This applies on the flop and before the flop. Therefore your multi size 3x answer won't do well against someone playing 3x only.
I have said that time and time again and it's an undeniable fact.
How do you study postflop? Do you keep pretending your opponent uses a multi sizing strat pre and therefore only has a narrow 30% ish range after raising the button?
Do still not see why this is potentially extremely dangerous to run sims with such wildly inaccurate assumptions?
As for the quoted part. What on earth are you on about? You act like the EV in poker comes from playing the most complex preflop sim possible.
i have more sizings preflop. therefore i will gain more EV than you. That's frankly the dumbest thing I have read on here so far. This isn't how poker works.
By the way in order to correctly multisize pre you would need to solve for 2 sizings pre. That's the way to do it. 2 sizings pre should work out fine with decent accuracy.
please enlighten me as to how that is possible. I'd love to reduce my RAM useage. I'm not saying this is impossible but what I am saying however is that in a youtube I have seen from Thomas he used 10 buckets per street and a low abstraction level. This has turned me off from their products entirely.
I don't think their process is good enough to answer "which sizing a solver prefers". I strongly believe we lack the technology to answer such complex questions.
Also if the strategies are not even designed to be used at the table what are they even for? Why would you study something you can't even use?
NGL aner0 you sound like you are a massive fanboy. Maybe take a step back and ask yourself if the zenith guys are potentially not telling you the truth. They run a business based on selling stuff no one else sells. All while claiming everyone else's work is bad.
I have said that time and time again and it's an undeniable fact.
How do you study postflop? Do you keep pretending your opponent uses a multi sizing strat pre and therefore only has a narrow 30% ish range after raising the button?
Do still not see why this is potentially extremely dangerous to run sims with such wildly inaccurate assumptions?
As for the quoted part. What on earth are you on about? You act like the EV in poker comes from playing the most complex preflop sim possible.
i have more sizings preflop. therefore i will gain more EV than you. That's frankly the dumbest thing I have read on here so far. This isn't how poker works.
By the way in order to correctly multisize pre you would need to solve for 2 sizings pre. That's the way to do it. 2 sizings pre should work out fine with decent accuracy.
please enlighten me as to how that is possible. I'd love to reduce my RAM useage. I'm not saying this is impossible but what I am saying however is that in a youtube I have seen from Thomas he used 10 buckets per street and a low abstraction level. This has turned me off from their products entirely.
I don't think their process is good enough to answer "which sizing a solver prefers". I strongly believe we lack the technology to answer such complex questions.
Also if the strategies are not even designed to be used at the table what are they even for? Why would you study something you can't even use?
NGL aner0 you sound like you are a massive fanboy. Maybe take a step back and ask yourself if the zenith guys are potentially not telling you the truth. They run a business based on selling stuff no one else sells. All while claiming everyone else's work is bad.
I'm definitely not a fanboy, I've been arguing with Thomas about a bunch of things including this one, I was on your side couple weeks ago until I saw the math and made the sims to see that he is right, Exploitability of 1-sizing sims is unacceptable and you can easily take advantage of it. Solver range EV is pretty much useless when you compare different trees. They're probably wrong about a bunch of things i don't even know or care, but they are right about how splitting sizings can exploit the **** out of pool specially if V tries to follow charts.
No human can split their range accurately into 5 sizings or whatever they recommend. Much like no human can split their range accurately into infinite sizes after the flop.
Even PIO or Monker won't do it very well. Because in order to make these huuuge trees fit you need to use heavy simplifications.
which reminds me... anero needs to tell me how to magically do this without sacrificing accuracy.
You don't have to apply the strategy as precisely as the other peeps, just err on the side of exploit. It's pretty easy.
I'm definitely not a fanboy, I've been arguing with Thomas about a bunch of things including this one, I was on your side couple weeks ago until I saw the math and made the sims to see that he is right, Exploitability of 1-sizing sims is unacceptable and you can easily take advantage of it. Solver range EV is pretty much useless when you compare different trees. They're probably wrong about a bunch of things i don't even know or care, but they are right about how splitting sizings can exploit the **** out of pool specially if V tries to follow charts.
I'm definitely not a fanboy, I've been arguing with Thomas about a bunch of things including this one, I was on your side couple weeks ago until I saw the math and made the sims to see that he is right, Exploitability of 1-sizing sims is unacceptable and you can easily take advantage of it. Solver range EV is pretty much useless when you compare different trees. They're probably wrong about a bunch of things i don't even know or care, but they are right about how splitting sizings can exploit the **** out of pool specially if V tries to follow charts.
Yes splitting sizes can exploit the **** out of people. That's also a fact. I also agree with some of the nonsense thomas called out (the no cold calling button stuff etc.)
I also agree that providing solutions only against 1 size is lazy for a premium product. At the minimum 3x and 2,5x should be covered. Just... again 1 at a time please.
What do i disagree with however very strongly is that's impossible to solve for 10 sizing split preflop in monker without sacrificing accuracy. It can't be done.
I can't imagine any settings that could possibly make this work.
Please do change my mind on this and tell me how i can do 8 RFI sizings, 8 3bets sizings and sufficient postflop sizings without blowing up the game tree?
No human can split their range accurately into 5 sizings or whatever they recommend. Much like no human can split their range accurately into infinite sizes after the flop.
Even PIO or Monker won't do it very well. Because in order to make these huuuge trees fit you need to use heavy simplifications.
which reminds me... anero needs to tell me how to magically do this without sacrificing accuracy.
You don't have to apply the strategy as precisely as the other peeps, just err on the side of exploit. It's pretty easy.
I'm definitely not a fanboy, I've been arguing with Thomas about a bunch of things including this one, I was on your side couple weeks ago until I saw the math and made the sims to see that he is right, Exploitability of 1-sizing sims is unacceptable and you can easily take advantage of it. Solver range EV is pretty much useless when you compare different trees. They're probably wrong about a bunch of things i don't even know or care, but they are right about how splitting sizings can exploit the **** out of pool specially if V tries to follow charts.
I'm definitely not a fanboy, I've been arguing with Thomas about a bunch of things including this one, I was on your side couple weeks ago until I saw the math and made the sims to see that he is right, Exploitability of 1-sizing sims is unacceptable and you can easily take advantage of it. Solver range EV is pretty much useless when you compare different trees. They're probably wrong about a bunch of things i don't even know or care, but they are right about how splitting sizings can exploit the **** out of pool specially if V tries to follow charts.
Yes splitting sizes can exploit the **** out of people. That's also a fact. I also agree with some of the nonsense thomas called out (the no cold calling button stuff etc.)
I also agree that providing solutions only against 1 size is lazy for a premium product. At the minimum 3x and 2,5x should be covered. Just... again 1 at a time please.
What do i disagree with however very strongly is that's possible to solve for 10 sizing split preflop in monker without sacrificing accuracy. It can't be done.
I can't imagine any settings that could possibly make this work.
Please do change my mind on this and tell me how i can do 8 RFI sizings, 8 3bets sizings and sufficient postflop sizings without blowing up the game tree?
Thomas must be working some magic here.
I'm sure you can backwards engineer it from their free stuff
Feel free to ignore my funny gif and just continue your conversation!
Feedback is used for internal purposes. LEARN MORE