So not sure if we can trust a program if we do not know how exactly it works.
I dont trust the software at all. It should be some approximation of nash equilibrium, but if you try playing vs their bots (who are trying to play "GTO"), you can crash them with pure aggression. And such bot should at least be unexploitable against aggression since it plays according to nash.
Originally posted by GingerKid
So not sure if we can trust a program if we do not know how exactly it works.
I dont trust the software at all. It should be some approximation of nash equilibrium, but if you try playing vs their bots (who are trying to play "GTO"), you can crash them with pure aggression. And such bot should at least be unexploitable against aggression since it plays according to nash.
These words of caution should always be kept in mind. It took a team of researchers at University of Alberta a decade to approximate GTO well enough for FLHU to call Cepheus "unbeatable" (whilst it is still an approximation). I am not confident at all that the required knowledge is in the hands of even extraordinary software developers in every case.
But I doubt as well that it is as easy as to "be aggressive" to beat the programs.
Originally posted by GingerKid
So not sure if we can trust a program if we do not know how exactly it works.
I dont trust the software at all. It should be some approximation of nash equilibrium, but if you try playing vs their bots (who are trying to play "GTO"), you can crash them with pure aggression. And such bot should at least be unexploitable against aggression since it plays according to nash.
These programs are VERY different than Snowie. And you can't "play" versus them, as far as I know.
But I doubt as well that it is as easy as to "be aggressive" to beat the programs.
I was talking about the snowie, not the HU bot from uni Alberta. 6 max is far more complex than HU, so it cant be compared.
Originally posted by YohanN7
Originally posted by GingerKid
So not sure if we can trust a program if we do not know how exactly it works.
I dont trust the software at all. It should be some approximation of nash equilibrium, but if you try playing vs their bots (who are trying to play "GTO"), you can crash them with pure aggression. And such bot should at least be unexploitable against aggression since it plays according to nash.
These words of caution should always be kept in mind. It took a team of researchers at University of Alberta a decade to approximate GTO well enough for FLHU to call Cepheus "unbeatable" (whilst it is still an approximation). I am not confident at all that the required knowledge is in the hands of even extraordinary software developers in every case.
But I doubt as well that it is as easy as to "be aggressive" to beat the programs.
Vorpal pointed me at this thread to see what I thought of it. I had a quick skim through it, and thought I should at least answer this point. I've written a GTO solver myself and, trust me, it's not so hard. My software agrees with Piosolver and GTORangebuilder in every spot. For me, the tricky thing about these solvers is not writing them, it's learning from them. I was given a copy of Piosolver and I can happily set up ranges and betting trees and compute solutions to my heart's content. I know lots of Pio users are aggregating results from multiple flops. Personally, whenever I use it I feel like a four year old with a calculator. I remember when my son was much younger he was holding a calculator and I asked him what 34 times 27 was. Click, click, clickety click.....123. Me: That must be wrong. Him: The calculator says 123! - stomps off.
I'm not a good enough player to have any use for my own software, which is sad, but there is lots of training material on the GTORB site, and Alex Sutherland's GTORB blog has lots of really good articles. He's a good player and a really understands game theory, which is more than can be said for a lot of poker players.
Originally posted by jbpatzer
Originally posted by YohanN7
Originally posted by GingerKid
So not sure if we can trust a program if we do not know how exactly it works.
I dont trust the software at all. It should be some approximation of nash equilibrium, but if you try playing vs their bots (who are trying to play "GTO"), you can crash them with pure aggression. And such bot should at least be unexploitable against aggression since it plays according to nash.
These words of caution should always be kept in mind. It took a team of researchers at University of Alberta a decade to approximate GTO well enough for FLHU to call Cepheus "unbeatable" (whilst it is still an approximation). I am not confident at all that the required knowledge is in the hands of even extraordinary software developers in every case.
But I doubt as well that it is as easy as to "be aggressive" to beat the programs.
Vorpal pointed me at this thread to see what I thought of it. I had a quick skim through it, and thought I should at least answer this point. I've written a GTO solver myself and, trust me, it's not so hard. My software agrees with Piosolver and GTORangebuilder in every spot. For me, the tricky thing about these solvers is not writing them, it's learning from them. I was given a copy of Piosolver and I can happily set up ranges and betting trees and compute solutions to my heart's content. I know lots of Pio users are aggregating results from multiple flops. Personally, whenever I use it I feel like a four year old with a calculator. I remember when my son was much younger he was holding a calculator and I asked him what 34 times 27 was. Click, click, clickety click.....123. Me: That must be wrong. Him: The calculator says 123! - stomps off.
I'm not a good enough player to have any use for my own software, which is sad, but there is lots of training material on the GTORB site, and Alex Sutherland's GTORB blog has lots of really good articles. He's a good player and a really understands game theory, which is more than can be said for a lot of poker players.
Hi,
Considering that today GTO for 6max no limit holdem is far away from being solved, what kind of GTO is your software doing? if you are providing approximation of GTO, then it should be said like that, and explain what are the assumptions about the approximated model. E.g. if you are building ranges for hero using GTO range builder, what is the assumption about villain range for particular board and actions? Are you taking in consideration all possible villain ranges (which should be actual GTO) or a single range?
First thing which I find unlogical, why is it for MP suggesting to fold around 44% range after checking the flop and CO bets 33% pot size? What does it have to do with GTO? It looks that the software has a huge leak in this spot. CO is checking back a hand like TT on the Flop, big mistake. It really needs protection, since there are many hands in MP range that are going to cf, like KQ, JQ, and similar, and have very nice equity vs TT. I agree that it is likely GTO to check back some value hands for CO, but those hands shouldnt be hands that need protection, it is usually high pairs, or even trips, set. Also, on flop CO is not checking back any strong hand (only some of TT, JJ which are very vulnerable), so it means that it can't handle big overbets once MP start betting turn and river after flop gets checked).
Thats why I wrote in a comment above, that I dont trust such softwares which just outputs "magic numbers" for each combo (EVs), and I should simply trust it that it does GTO magic.
Originally posted by jbpatzer
Originally posted by YohanN7
Originally posted by GingerKid
So not sure if we can trust a program if we do not know how exactly it works.
I dont trust the software at all. It should be some approximation of nash equilibrium, but if you try playing vs their bots (who are trying to play "GTO"), you can crash them with pure aggression. And such bot should at least be unexploitable against aggression since it plays according to nash.
These words of caution should always be kept in mind. It took a team of researchers at University of Alberta a decade to approximate GTO well enough for FLHU to call Cepheus "unbeatable" (whilst it is still an approximation). I am not confident at all that the required knowledge is in the hands of even extraordinary software developers in every case.
But I doubt as well that it is as easy as to "be aggressive" to beat the programs.
Vorpal pointed me at this thread to see what I thought of it. I had a quick skim through it, and thought I should at least answer this point. I've written a GTO solver myself and, trust me, it's not so hard. My software agrees with Piosolver and GTORangebuilder in every spot. For me, the tricky thing about these solvers is not writing them, it's learning from them. I was given a copy of Piosolver and I can happily set up ranges and betting trees and compute solutions to my heart's content. I know lots of Pio users are aggregating results from multiple flops. Personally, whenever I use it I feel like a four year old with a calculator. I remember when my son was much younger he was holding a calculator and I asked him what 34 times 27 was. Click, click, clickety click.....123. Me: That must be wrong. Him: The calculator says 123! - stomps off.
I'm not a good enough player to have any use for my own software, which is sad, but there is lots of training material on the GTORB site, and Alex Sutherland's GTORB blog has lots of really good articles. He's a good player and a really understands game theory, which is more than can be said for a lot of poker players.
Is there only one Nash equilibrium for given ranges for all possible bet sizes, and this equilibrium is stable ?
Post I've found from the creator of GTORB on how it works/what it's useful for
Here
Good read, very informative video.
Originally posted by LemOn36
Post I've found from the creator of GTORB on how it works/what it's useful for
HereGood read, very informative video.
Ok, the blog looks good, thanks for the info. Can some of the coaches please say opinion about quality of such softwares? E.g. about flop ranges in the video posted in this thread? why is it having so big leaks (e.g. MP folding 44% of range after checking flop vs 33% bet size)
What makes you think that this is a leak with the ranges both players get to the flop with?
Originally posted by LemOn36
What makes you think that this is a leak with the ranges both players get to the flop with?
If MP checks, and CO bets 33% pot size, then CO needs around 25% FE for autoprofit for bluffs that have 0% equity. That means that MP should fold max 25%, to prevent CO have autoprofit. Since almost all bluffs nearly always have some equity (decent equity) on flop in this spot, then MP needs to fold even less than 25% to prevent autoprofit. Since in video, MP folds 44% vs 33% pot size, it is clear that CO should bet all bluffs and exploite MP this way (CO is not betting all bluffs in this spot, which is also a mistake). So since CO can increase his EV by betting all bluffs on flop, it means that equilibrium is not reached at all! Equilibrium is only reached, if none of the players can change their strategy in a way to increase their EV. If MP would have unexploitable ranges (and reached equilibrium) then CO wouldn't be able to change his strategy and increase his EV.
The mistake of folding 44% on flop is even bigger because MP is checking most of the time (it wouldn't be such a problem if MP would cbet flop e.g. 70% and CO has autoprofit only 30% of time). If the reason why MP is overfolding so much when checking flop, is because CO has position and better range, then MP should check the whole range, in order to minimize the EV of CO (because of the very similar reason as when we cold call preflop we usually check whole range to RFI because he has range advantage).
I think it's been widely accepted now that 1-A defend necessity doesn't really apply in loads of spots, and it's okay to leave villain's bluffs to autoprofit
And from what I gathered from players that look at GTO and even great HU players, this is one of them, where you should be "overfolding" flop and it'll be the correct strategy.
So since CO can increase his EV by betting all bluffs on flop, it means that equilibrium is not reached at all! Equilibrium is only reached, if none of the players can change their strategy in a way to increase their EV. If MP would have unexploitable ranges (and reached equilibrium) then CO wouldn't be able to change his strategy and increase his EV.
This I am curious about 
I think it's been widely accepted now that 1-A defend necessity doesn't really apply in loads of spots, and it's okay to leave villain's bluffs to autoprofit
And from what I gathered from players that look at GTO and even great HU players, this is one of them, where you should be "overfolding" flop and it'll be the correct strategy.
Are you now talking about exploitative play or GTO? Yes, even in GTO (unexploitable play), there are spots where we have to overfold, in spot where villain has stronger range. If that is the case in the scenario from video, then MP should check the whole range, to minimize the EV of CO's range, and that way loose minimum vs stronger range. I didn't make calculations which range is stronger in this spot, as I noticed it is also not mentioned in the video, but usually RFI (MP in this case) has significant range advantage on low boards (especially on dry boards), and should with such range advantage cbet with high frequency. So my point is, what ever the case is (MP has stronger range, or CO has stronger range, or they have equally strong ranges) the ranges constructed on flop don't have much to do with GTO and especially not with reached equilibrium.
I can tell you that, given the ranges and the betsizes allowed, that solution is very close to a Nash equilibrium. It is difficult to explain how CO looses EV by putting more bluffs in his bet 33% range. I have to assume the EV he gains there he looses in the other game trees.
Tackleberry is one guy that could shed some light on this. I'll try to get him here. Sorry that I don't have a better answer. Anyways, at least we are discussing some interesting stuff
The solutions are at equilibrium and don't have leaks (there are limitations, but these are mostly insigificant). As jb says, check out asuths stuff at GTORB.
I can tell you that, given the ranges and the betsizes allowed, that solution is very close to a Nash equilibrium. It is difficult to explain how CO looses EV by putting more bluffs in his bet 33% range. I have to assume the EV he gains there he looses in the other game trees.
Hi,
CO needs around 25% FE with 0% equity bluffs to have autoprofit, and MP offers him 44% FE. I can't see a better spot than this to maximize EV. If there is some other branch in the tree, where CO gains even more EV, it is also even a bigger leak from MP. I would like to hear some explanations, why is it equilibrium, would be nice to hear it from Tackleberry.
Originally posted by jules97
The solutions are at equilibrium and don't have leaks (there are limitations, but these are mostly insigificant). As jb says, check out asuths stuff at GTORB.
Theory is one thing. Implementation of a software is other thing. So even if theory is 100% correct, it doesn't mean that in practice the software will work 100% correct.
GingerKid: you dont want target goal, makes villian bluf to 0EV, but makes indeferent between bet and check... You cant simply (IMO) look on the flop and say "I need X% Feq because I bet Y%, and if villian doesnt defend 1-a, then he made mistake" Because game started on previously street, where for example villian invest money to the pot (for example he call open raise -> he get some pot odds, which given how often you must win our hand) -> from which it derives -> therefore (imo) looking for example on the turn and using aproxiem. (1-a) and wondering, why snowier or this soft "overfolding" is simply bad approach.
Sry for bad en.
but honestly I dont understand why snowie for example in his preflop advisor, advise 3bet this range BBvsBT(2.5bb) and when I choose I am facing a 4bet, then he defend only this range
its about ~67% feq
Originally posted by HHHuBi
GingerKid: you dont want target goal, makes villian bluf to 0EV, but makes indeferent between bet and check... You cant simply (IMO) look on the flop and say "I need X% Feq because I bet Y%, and if villian doesnt defend 1-a, then he made mistake" Because game started on previously street, where for example villian invest money to the pot (for example he call open raise -> he get some pot odds, which given how often you must win our hand) -> from which it derives -> therefore (imo) looking for example on the turn and using aproxiem. (1-a) and wondering, why snowier or this soft "overfolding" is simply bad approach.
Sry for bad en.
Sure, I agree totally about "1-a" and making villain indifferent, thats the basics, but still it can't be : "CO needs around 25% FE with 0% equity bluffs to have autoprofit, and MP offers him 44% FE. "
You say, MP offers so much FE to CO because this way CO is indifferent for checking or bluffing, but this is only possible if MP range is so much weaker (and thus his check range must be very weak), which is not the case here. Even if it would be the case, then the solution would be to check the whole range (also basic thing, if our range is weaker, we want to check most of our range regardless of preflop initiative).
