Toy Gaming (part 3)
Posted by Sauce123
Video Player is loading.
This is a modal window.
The media could not be loaded, either because the server or network failed or because the format is not supported.
Resume Video
Start from Beginning
Watch Video
10
Loading 45 Comments...
wow first time in my life im writing it but i seriously have to when getting a chance esp. in a sauce vid .... *FIRST :p
Hey Ben,
good demonstration -- i doubt there will be many Qs on the models, i thought you were very clear in your points.
i'm going to paste a quote from you late in the video and then try to relate it to what i was getting at in the thread. Hopefully more discussion can advance from there.
28min - "we bluff in order to get action on our value hands, not to get value when we bluff"
I feel like this is a huge point and common misconception.. especially with regard to overbetting, where generally i find there is a ton of confusion that springs up about the larger sizings functioning exploitatively. Cruising your earlier videos, i noticed you faced a lot of overbet questions from members like "i feel like the overbet is good exploitatively as a bluff, but i can't see how it serves a balanced range". Confusion should be cleared up from that quote.
ITT, I was interested in the possibility of taking a sub EV line with a portion of our value range to unilaterally increase another portion of our value range (no bluffs in this example).
The best example for my question is the 872 flop from Tyler's CREV analsysis vid. MP v BTN, you and Tyler both seem to like the idea of OOP PFR devising his strategy on the flop in a way that allows him to realize more EV with overcards.
It seems like if we wanted to specifically maximize the EV of our overcards, there is no better strategy we could take to accomplish this than checking our whole range OTF, since IP must check back most often vs that strategy. My concern was how much EV we sacrifice with parts of our premium value range when villain checks back at a high frequency and takes a street out of the game.
This left me wondering if it was ever possible that we would be justified in sacrificing some EV with some premiums OTF in order to help overcards retain more equity (since we have so many overcards). The general consensus from the authorities aka GT was no, which makes sense. But from a practical sense, it's a massive energy saver as a grinder to narrow my strategic options OTF, which makes checking range very appealing.
Thanks for any thoughts on this -
Nick
The idea is that if it's correct to check 100% of our nutty hands on the flop, then the EV of checking them will be greater than or equal to the EV of betting them. If that's not true (and I agree with you that it's plausible that it isn't in the 872r example), then we need to begin betting a % of our nutty hands until we're indifferent again. Tyler and I agreed that if we do need to bet our nutty hands it probably isn't a very large frequency of them, because once we weaken our checking range on the flop IP can quickly begin to bet the flop with a high frequency denying us the substantial amount of EV we have with our overcard combos vs a check. If we chose to look really closely at this situation, I'm sure we could find the right mix of nutty hands to bet, but the problem is that once we spend the time simulating all the various runouts and testing various assumptions about villain's play on all of them, we've created a model which isn't particularly generalizeable to other situations since it's so sensitive to our assumptions.
The upside to this discussion is that if we do need to take a mixed strategy with our nutty hands, then it's also true that we won't lose very much EV by making the mistake of checking all of them. That's because for every bet we lose with a nutty hand, we gain substantially with our overcards and weaker hands, and we'll hold many more of those combos. Because of the robustness of checking nut hands on this flop, we can feel pretty good about our model once we identify that strong hands need to be checked at a high frequency, and unless we're really curious we don't have to go any further.
Things are actually a bit more complicated in this case because on the 872r the difference between betting and checking overcards was effected substantially by how villain plays both preflop and on the turn and river. So, really, we're just kind of doomed on this flop whether we bet a lot or check a lot because our distribution contains so many overcards that we can't possibly realize our equity without giving up something. Blocking bets might be the choice too if we can figure out a way to get substantially more cheap cards with them.
Anyways, to move back to the theoretical point. If the EV of our game tree is higher given we make some -EV bluffs, then either a) We've picked a dominated betsize (e.g., the larger bets in the AKQ game for IP), or b) We haven't figured out the right mixed strategies for each player to take such that we're at equilibrium. But we're never at equilibrium if we're making -EV bets.
The point you made it is possible to sacrifice EV in a spot to make a future spot with higher EV than the natural GTO play later. It isn't for some assumptions that GT are based that aren't quite explained in MoP.
Standard GT assumes that information is complete. The implication of this assumption is that you know your opponent perfectly well and all history of the hands you played is equal in the heads of both (and all your opponents also have access to this information). Besides that, all the levelling (to the highest possible number you can think of) is done in both heads. In addition to it, to standard GT the world ends after the game is played. There is no sequential game to the one played. When you play another hand with the same player the information set would be roughly the same of it was before, considering that you have enough hands to make the information set on the player the wider it can be.
This fundamentals of GT make its application limited to certain streets and are better used OTR. If you play a guy 1000 hands HUNL you want to maximize the payoff of the session. GT doesn't consider the whole picture and can be used to focus only on one street at the time (for sequential/repeated games, the equilibrium won't be found on the same way). If you are balanced and playing each hand on the proper GT fashion your expected value will be higher (or equal) than any other possible strategy.
There are another type of GT analysis, that is Bayesian GT which the aim is to find a Nash equilibria for games that information isn't complete, like poker. After SCOOP I may make a thread on theory about it and how it can be used to improve the logical thinking on poker/street projection.
You guys probably noticed but english isn't my native language, and any doubt can be asked.
ps: toy gaming 2 is the best thing ever happened to poker since RiO
These statements are not entirely correct. We bluff because we try to make money on our bluffs. Or to be more precise, to make the most value out of our entire range.
To make this more formal:
"We have bluffs in our betting range in order to make money on our value hands when our opponent is playing a perfect response to our strategy. It is not the case that we bet our value hands in order to have profitable bluff hands when our opponent is playing a perfect response to our strategy."
The second part of the statement is only partially true. With our value hands, the only way to make more money is to bet and hope to get called. But consider the following example, the board is 4c5c8c9hTh and our value range consists of nutflushes. If we would only bet our value range, then we only get action from the straightflush, so we never get value, hence we shouldn't bet at all.
Now suppose we start using our 9 combos of nutflushes, Acc, and 3 combos of bluffs, Ac2!cc. Then every flush will face 7 value combos and 3 bluff combos. So if we bet pot, every hand but a straightflush has to fold as every bluffcatcher will have between 25% and 30% equity. Now all the Ac2!cc combos have a profitable bluff because they can claim the entire pot!
Clearly we bet our value hands so that we can bluff, in this game our value comes from bluffing only.
Recall that this strategy is not optimal, only an improvement. In fact, since our opponent is always folding when we bet, our value hands prefer to check since they only get action from the straightflush!
GT,
wouldn't we just include more bluffs other than Ac2!cc until we break even on bluffs and get max value from our nut flushes? Isnt this the play that "maximizes the value of our range"? and that is accomplished by bluffing enough such that we get called by our value range? Perhaps I'm missing something from a point you are trying to make.
@ZF
Yes, but that just complicates the situation. The problem with these statements "we do X to accomplish Y" is that they apply to a situation where we have a certain strategy and our opponent is playing a perfect response, and then we change this strategy by doing X, and accomplish Y after our opponent is playing the perfect response to our new strategy. This first situation is almost never a Nash equilibrium, otherwise these statements become meaningless.
You could imagine the same situation, where betting with the naked Ks becomes a break even bluff and our naked As bluffs have positive expectation. Clearly we like to have bluffs with a positive expectation, and betting our strong hands allows us to do so. My example was just one of the easiest examples that I could come up with that has this property.
I agree. The true statement is we try to maximize the value of our range. But to someone who isn't fluent in game theory terminology that can come off sounding like a tautology and so won't teach them anything. Of the two not entirely true statements I think saying that we bluff so that we can get value on our good hands gets people into less trouble.
It depends on your preferences. The answer is that in a cashgame you must do both, bluffing and value betting.
The way you think about it is that betting with value hands is 'natural', if you have a good hand you might as well bet, maybe you get called by worse. So you implicitly assume that people are calling with worse for some reason. And you leave out the reason why they call with worse hands when playing a perfect response to our strategy. This reason is of course because we are bluffing!
I think that my example shows that betting a value hand is not always 'natural'.
The downside of the AKQ game is that when you hold the A you have the absolute nuts so betting becomes weakly dominant, whereas in most forms of poker our value betting range is much wider than the absolute nuts. So betting those hands is no longer a (weakly) dominant strategy. Thus a part of this trouble that you want less people to get into comes from the choice of your toy game.
I don't really understand the practical application of these videos. I get that if we bluff with QQ then villain should call with KK some % of the time, but how does this help in real world games? The toy gaming videos are a little to dry for my taste. My favorite ones are the Heads up games where you go in depth about a hand and we can see your real world adjustments on the fly.
Thanks for the comment. I'm going to try and do a better job applying these toy games to real poker situations in the future. I think it takes some imagination to really see the connection between the toy game and adjustments to make at the table, and if you guys aren't making those connections I need to do a better job as a video maker.
Let me give you one typical poker example. We're playing 6m nlhe and we raise the btn with 87dd and get called by the BB, flop JcJc5d we cb and get called. Turn is a 6d and we barrel and get called, and river is a 2s. Suppose that going into the river villain's range is 1/3 Jx+ and 2/3 88< (sounds pretty plausible to me). We've stumbled onto a spot that's a lot like the AKQ game in many ways. We'll probably want to use a betsize of around 50-80% pot, since given we hold a J, his calling range will be less than half J>, and given we hold a bluff, we'll insure that we don't tighten his calling range to include mostly Jx.
Ben,
Suppose you took a good chess player who intuitively understood the concept of equilibrium and you decided to teach them NLHE. What would be the order of situations or resources that you would teach them to master? For example, lately I've been thinking that this may be the best way to build poker from the ground up:
1) Understanding the optimal strategy for poker with no blinds (Trivial)
2) Chapters 11+ of MOP
3) HU pre-flop play in both Limit and No-Limit
3) Three-handed preflop play (BTN vs SB vs BB) for No-Limit
4) Six-Handed preflop play for No-Limit
5) Toy gaming postflop models and understanding how these models deviate from reality by running equity simulations and understanding equity distribution/cards which favor one range versus another
6) More in-depth models of bet-sizing and constructing mixed strategies post-flop
7) Exploiting weak strategies and finding the Nemesis versus common mistakes
The best resources I'm familiar with are:
MOP
Flopzilla
CREV (I have not used it but have seen demonstrations on here and it seems exceedingly useful)
EquiLab
HM2 (For mostly exploit stuff)
Can you think of any other resources or would you change anything about the sequencing above?
Thanks,
Adam
Great question, but unfortunately I don't have even the beginnings of a good response.
I've found that modelling situations in CREV and looking at toy games are my two preferred ways of studying. But, to be honest, #1 by far is and always will be playing tons of hands against tough players.
Thanks Ben.
I've heard this a lot, but I find it hard to learn by playing without review.
Are CREV and HM2 your preferred methods of learning from the hands against tough players, or are there other ones you find useful?
Also, you said:
But, to be honest, #1 by far is and always will be playing tons of hands against tough players.
So would you throw game selection out the window if it meant moving down in stakes for awhile? Did you practice any game selection when coming up through the ranks?
I'm torn on this, because on the one hand it seems that finding the toughest players to play against isn't all that necessary when work away from the tables could yield improvement against complex strategies without costing you for your mistakes, but on the other hand there's a definite complacency that comes with winning every day from playing with the biggest morons you can find.
I'm not sure if it's possible to get really good at poker playing droolers, maybe it is if you study really hard.
I know I couldn't be very happy if all I did was play morons, it's the competition of poker that gets me excited to play every day (within reason). Now, that being said, I'm one of the few people lucky enough to be able to play a fair amount of tough competition and still make a very comfortable living, and I'm super grateful for that, and I don't think that's a realistic goal for every player. But I think that even if you don't plan on trying to beat highstakes, it's good for your heart to challenge a boss player at your limits once in awhile if only to get you fired up to improve your game and test yourself.
Hey Ben,
I think you're right. I've heard too many first-tier players say this about what helped them improve.
It's kind of the difference between yours/Galfond's approach to poker as competition, and the Leatherass approach of poker as a business. I've often felt more congruent with firing up 12-16 tables to play with whoever I see than sitting around waiting for droolers. It also alters my play to the point where I start playing not to lose rather than to win.
I've actually started viewing players better than me as my "coaches" at the table who show me what I need to improve on.
In the Nuts vs Air toy game the optimal betsize was all in. In this game the optimal betsize is 0.41 pot. Why is the result between these two games so different?
that was nuts or air against a bluff catcher. in the AKQ game your opponent can have the nuts in addition to his bluff catchers.
Any plans to work something like this out across multiple streets in the future? I think that would be really helpful for me personally.
Yeah some multistreet game theory videos/concepts are highly needed.
Hi Ben!
always love your vids!
i think i maybe found the 'bug' where you talk about on 14:30. The flop you use for this toy game is better for AA and QQ vs KK. because it can run out quads deuces. or a 2 and A/K when it is QQ vs KK for a split. I think when you will use a flop like 26J rainbow this 'bug' will be solved :D because then AA and QQ have 50% chance vs KK.
I think I input a known turn and river card to avoid this, but I might have forgot, thanks for the heads up.
hey Ben,
Thanks for the video. Why haven't you found "r" (MOP "r") useful in actual NL holdem? Can you think of any actual example where it would be best to bet around 41% pot IP on the river? It seems like it's not rare that I find myself IP OTR with what I'm pretty sure is the best hand but I don't feel awesome about betting it because villain can hardly ever call. Would that potentially be a good spot to bet "r", along with any bluffs I may have gotten there with? Is that too vague of a question?
If you think villain can't call you very often then it's probably right to bluff more and continue to value bet big with your strong hands. If you think you can't get called by worse then it's time to decrease your betsize.
damn when did RIO sign James Harden to do vids?
Ben, you mentioned above you were planning some multistreet scenarios which I'm very much looking forward to seeing. One situation I was wondering about is where the multistreet action has narrowed ranges to an extent where our opponent will hold a blocker to our value range a very significant amount.
When we create our river betting range we add the appropriate number of bluffs to our vbets, however assume our opponent is clairvoyant to our strategy, then he could make too many +EV calls within his range for our bluffs to be neutral EV (his blocker would weight us to bluffs given our true range and he contains it too often due to board texture and action). One option could be to reduce our betsize such that our opponents normal folding region doesn't contain the blocker anymore and thus he can't make too many +EV calls. However, assuming we want to bet large (say our value range is the nuts) then how would we solve this problem? Would we need to reduce our actual bluff frequency so that a larger portion of his range is truly indifferent to calling our river bet? Or do you think that this situation should never arise in optimal play as its unlikely our opponents range should contain such a blocker too frequently/our value range should be wider? This situation has come up a couple times in PLO for me, although rarely. Any thoughts would be much appreciated.
This sort of question is more appropriate for a video than for a comment thread. I'll keep it in mind.
Very interesting topic.
It was discussed here : A Deeper Look Into Blockers
if we are facing a river bet against a "GTO player" (at least someone
trying to make us indifferent to calling or folding, and at least in a
perfectly polarized vs bluffcatcher situation), then
- whenever
we have a blocker to the bluffing combos (like in the JdJx first hand),
we know that the bettor is unknowingly unbalanced towards value hands,
and we have an easy fold, because we cannot be indifferent to calling or
folding : calling has a negative expectation instead of the constructed
zero EV.
- whenever we have a blocker to the value combos, we
know that the bettor is unknowingly unbalanced towards bluffing hands,
and we have a mandatory call, because our expectation is always positive
! In this case, the bettor still has a positive expectation, but lower
than the constructed 1 pot EV.
Sounds like a huge implication of blocker usage.
I would love a concept video about it.
what if when we have AA and villain can just have KK or QQ. I guess here we should bet as much as we can but allways check QQ some small % as it is money in the pot? or am a I completely wrong here?
I think white devil answered doncamatic and your question a few posts above
if when we have AA we cant tie and its just 1 combo of AA
Nice video! I believe that you are a genius and that beard you are growing makes you also look like one. Love it!
I don't know how to quantify this, but would showing opponents bluffs to tilt them/pick up later EV be the same as giving up some EV to improve our value range? Showing a bluff or saying something in chat doesn't seem +EV immediately, but possibly can increase EV later.
Hey Ben, sorry if im some kind of late on this video.
I have a quick and simple answer if you still could read it. How do you run those graphs where you check the size that maximizes the time of OOP calls with KK ?
Awesome video, and hoping for a lot of more toy games !! Thanks a lot.
Hi ben, did you put the CREV file anywhere? I didnt find it in the thread. TY
Lol this vid is funny if you assume its a three handed game
Hi sacue, this video is too good, it should be illegal. Anyway I have a question.
I find it easy and extremely useful to apply bluff ratio alpha in many spots during my everyday's grind, but when it comes to calling the way that will make IP indifferent to bluffing - I'm lost.
Its logical for me that I want to arrive at such calling frequency that IP will not have autoprofitable bluffs. So what I would do is just try to cut 1-alpha of my total range and make sure that all of these hands beat a bluff. Now I've been told by you to do different thing: cut 1-aplha of that part of my range that beats a bluff. You asked in a video to think about it for a minute, I gave it half an hour yesterday, another half an hour today - still have no idea how can that make sense.
Let's use random example where from my point of view this method fails - I arrive OOP on the river with such range that I have 10 combos of value hands, 10 combos of bluff catchers, 20 combos of non sdv missed draws and we face a pot siezed bet. Applying your rule I end up folding 75% of the time (1-alpha of my 'beat a bluff region') and my opponent have incentive to bluff more often.
Why not use all our bluff catchers (so 1-alpha of my total range) in my example and stay unexploitable? Where am I missing something.
I play HU SNG hypers and there's a lot of shoving going on on the river in single raised pots and we build sample in common spots very fast. Having well balanced call off ranges that I build using strong theoretical concepts that I fully understand would be so extremely useful, so thanks in advance for any help.
pretty sure i love you ben
good grief, who is that hirsute yeti pretending to be Sauce?
Be the first to add a comment
You must upgrade your account to leave a comment.