As of November 2021, the APT Forum is closed to new posts. Like with many online forums, usage has decreased in recent years. All previous posts are still available.
Sklansky Fundamental Theorem Of Poker - Limitations
I think that in today's game Sklansky's theorem has some limitations. For example, let's say that you have AA and I have JJ. However, if I am 100% certain that if I flop a set, you will go for at least 20 times the pre-flop call, I think that making that call is correct even though the theorem would call the pre-flop call an error.
It appears that the Bots don't fully take that tendency into account (or perhaps don't give it the proper weighting).
Sign In to comment.
Welcome
It looks like you're new here. If you want to get involved, click one of these buttons!
Comments
But aren't you still saying that you'd be playing the hand exactly as you would if the cards were face up?
In terms of the formula used to calculate IQ points, it sounds like that works pretty much the same way that money does in real life. You lose most of the time when you don't flop a set, but when you do flop the set, you make enough profit to offset the losses in the long run.
Apt_gs,
Nytider is correct - if you overall are making the same play you would if the cards were face up (and you were correct with that play), you will be winning in accordance with the Sklansky definition. Now if you are incorrect and you cannot profit the way you think you will, then there is an underlying problem with the way you are playing and at least in our game, the IQ score will take that into account in the long run.
I think the one criticism you could bring agains the Sklansky theory is that sometimes people would play the cards the same way if they were face up and that way is incorrect in terms of odds. Our IQ takes that into account, but by the literal definition theory of poker, that isn't really considered as a possibility.
Thank you both for you comments. I think that you have helped me understand the concept better.
From having looked at IQ score, I'm not convinced that, however it is formulated, that it accurately accounts for the implied odds apt_gs suggests. I would also say that I'm not convinced that it is particularly useful as a measure of how one plays inasmuch as one is never (or almost never) making decisions based on putting an opponent on one hand, but on a range of hands.
I believe IQ score will tell you that you are never 100% correct to call even a small preflop bet when your opponent holds A-A, even if you hold K-K. But if you are playing against your typical LAG, who you know could hold a wide range of hands, you're 100% correct to play K-K to the max.
It doesn't make a lot of sense to me how one can have almost idealized frequencies in the "skills" metric (how often did you see the flop?, how often did you raise pre-flop?, how aggressive are you?, etc.) and have an IQ score around 100.
I think, even if you played the professional adviser recommended strategies 100% of the time, you probably wouldn't have a session IQ score above about 110.
Thanks for the comments MAM4. Mathematically, if you are correctly estimating your opponents' hand ranges, the IQ score will be accurate in the long run. It takes that into account, as well as how good you are at estimating implied odds. Because if you are correct about the implied odds, you'll win an abnormal amount from your opponents.
If you were to take the advisor recommendations all the time, your IQ score will be somewhere between about 90-100 because it comes down to computer vs. computer. But humans are better than these computers, so you'll be slightly below the average human.
Are you playing in KGB or in a different level game?
I'm playing KGB level. Just to confirm, I ran a "Doug Hull Adviser" simulation where I played 500 hands exactly to the adviser's recommendations. "Doug Hull Adviser" had an IQ score of 94 and got crushed, losing $1331 for the session (and putting a dent in my positive BB/100 average - LOL).
I guess my point would be that I see no way that any player, no matter how good, could generate a long term IQ score over, maybe, 110-115. You're always going to make a bunch of "mistakes" per the Skalansky theorem (because you have very imperfect knowledge) even if you are playing in the maximum feasible way to give yourself the best return.
Hypothetical question - Let's say you played an infinite number of hands holding A-x suited, get in preflop in a multiway pot for no more than a 3 BB raise, and flop the nut flush draw on an unpaired board. Then every time you get it all in on flop for 3.5 to 1 pot odds where your about 3 to 1 to hit the flush and win the hand. You go on to lose 2 out of 3 hands, but you make a net profit because you are correctly getting it in with positive equity odds. What would IQ score judge your play as?
I wonder how that would end up playing the Daylian Cain advisor? My personal favorite.
Syn
I'm sure it would do just as badly.
Which raises the question - if the best the advisers do is an IQ score in the 90s, why the heck would I bother paying any attention to the advisers when I can outplay them already?
I would also point out that another thing the Skalansky theorem doesn't capture is the old saying that "poker is a game of human played with cards." The point being that your objective isn't necessarily to make the fewest "mistakes." It is to play in such a way that you cause your opponents to make more and more costly mistakes than you do.
MAM4 et al - Thank you for your comments. They have definitely given me a lot to thank about. I think that having a forum where thinking poker players can exchange their thoughts is a very positive thing for my game.
MAM4 - I think you might be misunderstanding how the IQ is calculated. It takes into account the "playing the human" that you talk about. Here's a link to a detailed post I made describing it:
https://www.advancedpokertraining.com/poker/forum/discussion/comment/145#Comment_145
As far as the bots - they are computer vs. computer - they by definition can't be better than average because they are playing against themselves. Their advice is good for something to think about - I would definitely not recommend taking it because then their play will influence your training plan, but use the brain button to see why they thought it was the right move. Study the estimates of the opponents' ranges and see if you agree with it. Look at the odds against the ranges, etc. I think that is the real benefit of the game and why the site really works well - over time you learn the odds, how to estimate ranges, ways that other people might potentially play (since the computers are designed to play more like humans), etc. The bots are NOT designed to be perfect GTO computers, because humans don't play that way. I think it is true that many of our users can outplay the bots at the lower levels - but not at KGB. There are actually very few people who are winning over a large sample of hands in KGB's dungeon. Those people who are winning are typically crushing it, and they are the ones getting the super high IQ scores.
What is "KGB's dungeon"????
KGB's dungeon is the highest level game on the APT site. It is the one with the best bots, and any peeking at opponent cards is not allowed. It is the best test of skill available on our site.
its "Final Table Trainer" ???
No, it is a game level on the 9-max or 6-max cash game. When you start the game, you can choose the table - the KGB's Dungeon table is the hardest level.
thanks
Is the Las Vegas Main Event the MTT equivalent of KGB Dungeon? Do the AI opponents play at the same difficult level?
Yes, it is the equivalent. There is a little more variety in the opponents, but the overall ability level is the highest in that tourney.
Allen,
Let me ask you a separate question. In terms of the IQ score, what would you expect it to roughly be for a player who over the long term is, say, +2-3 BB/hour playing $2-$5 NLH live at a Las Vegas casino?
And, to be clear, I'm not saying there is not a lot to like about the site. I just put a lot more stock in what my BB/100 is playing a particular level and what some of my other peripheral statistics are vs. my IQ score.
I just think that because the bots don't adjust to how you play (unlike human opponents, at least the better ones) the whole IQ score/Sklansky theorem may push one toward a much more bifurcated range of plays (like always limp with small pocket pairs preflop) which is more easily readable by actual human opponents.
The IQ score is completely for fun. There is basically no training value in it. It is extremely accurate (perfectly accurate in fact) in the long run, but very noisy in the short run. You'll get a lot more from the training plans, studying opponent ranges, odds, etc. I'd expect the long-run IQ score for a player like you described to be high compared to the opponents he is playing. However, compared to everyone using our site, it is tough to say. The IQ score is relative to all people using our site, not relative to everyone who plays poker.
The bots DO adjust to how you play in the top level game. They use AI and learn to beat you based on your tendencies. That's why very few members (<10%) beat the top bots in the long run. The ones who do typically have discovered a weakness in the bots, and they do have them. We are always working on plugging their leaks as we find them, but I know some exist that are difficult to plug without causing new exploitable weaknesses.
Allen,
Thanks. You make a couple of interesting points here.
"The IQ score is relative to all people using our site, not relative to everyone who plays poker." - That certainly changes the way one might interpret the meanings of the various IQ ranges. I would assume that the people that self-select to subscribe to a site like this are already typically among the top 50% (at least) of the population of 1-2 and 2-5 players. So being "above average" for this site probably puts one even further above average for the total population of 1-2 and 2-5 players.
"The bots DO adjust to how you play in the top level game." - Also good to know. I'm pleased to hear that that capability is programmed into the bots. When you say "in the top level game" do you mean that is only a feature at KGB level? In the limited amount of hands that I've played since I came back to practice, the bots do seem to be a little susceptible to being double barreled or float-bluffed off hands.
The learning isn't limited to KGB or top level tournaments, but the bots at the top level use the information much more effectively. So if you do things like double barrel bluffs, they will pick up on that and top ones are more likely to punish you for it. How exactly they do it, I couldn't explain. Steve and our other programmer Gabriel have done 100% of the work on the bots and I think they are the only two people who really know exactly how the software learns and uses the info. But I know from experience using the site and talking with them about the design that you have to switch things up or the bots figure you out.
I have definitely noticed that a float-bluff line is reasonably effective, even at the "hardest" level. I can't really ay how it compares to the effectiveness of the same strategy in the wild. But I absolutely can attest to the bots learning, or at least mixing up their response to this line. Here is an example that happened to me the other day in the MTT simulator set to the hardest level:
I flat call in position pre-flop. The bot, who was the aggressor on the previous street, makes a C-bet. I call with air. The turn doesn't help either player, presumably, and the bot now checks to me. I bet about half the pot, and the bot folds. I do this a couple of times over the course of the tournament.
Then, all of a sudden, the same scenario plays out, except that instead of check-folding on the turn, the bot check-raises. This, of course, could just be random. It could be that I just ran into a bot with a great hand. But it sure had the feel of the bot having "learned" what I was doing and then having taken appropriate action to exploit it.
But again, in terms of translating this from the AI world to real life, I think there is a strong case to be made that this is exactly what real opponents would do. I know it is exactly the type of play I'd make on somebody who went to the well once too often with the bluff on the turn.
Question: Under Sklansky's theorem, is this a mistake all of the time, some of the time, or none of the time?
The board is As 10h 6d. Player A has 77 and bets. I have 44. At this point I am beat and I have a very small % chance of improving to the best hand. However, I feel that if I raise, there is a 80% chance that player A will fold. If I simply call, given that I am beat now, my equity in the pot bet is -EV . However, if I raise my equity is +EV.
So, would Sklansky's theorem say that:
1) Anytime I do anything other than fold I am making a mistake (since I am beat at the moment of A's bet).
2) The 80% of the time that A folds (with a better hand) he is making a Sklansky "mistake", and the 20% of the time that A calls, I am making a Sklansky "mistake"?
3) 100% of the time that I raise, the play is correct given my overall +EV
My guess would be number two, since the player with 77 would never fold if the cards were face up.
I don't think the individual situation matters - it is your estimate of the range of your opponent and your opponent's action that is what matters in the long run. If your interpretation of the odds of folding is correct and you know for certain your opponent has 77, then the play is correct. It's the same basic thing as the KK vs. AA I mentioned before in some thread - you don't know what your opponent has. So whether your play is correct is a join function of your estimate of your opponents' range, and your estimate of what your opponent will do (and of course the pot). If your estimates are correct, on average you will make money. But any time your opponent is able to get you to misestimate their hand and do something wrong, or you just make a mistake in thining about what your opponent will do, you will lose money.
I think this video is a very interesting play out of this if you haven't seen it yet. Unfortunately, I can't find the full video that shows Esfandiari talking with Negreanu about the exact situation right before this hand. He said he has only folded KK five times in his life.
https://www.pokernews.com/news/2017/08/tom-dwan-antonio-esfandiari-poker-after-dark-28756.htm
From a pure technical perspective on Sklansky's theorem, that is correct.
Thanks to both of you. My guess was also #2, but Allen's explanation really helped to solidify the concept in my mind. Also thanks for the link.
I had the full video bookmarked, but I don't know what happened to it. I can't find it now either.It was very good though. They had a discussion right before this hand about folding KK pre-flop. Daniel said, if I remember, that he had done it twice. He was right once and wrong once. Antonio claimed five times, but "years ago." He indicated that he would not do it now.
Yep. So I think I'm going to write a blog post about this and my thoughts about it. I'll link to it when I'm done. Very interesting situation.
Here's another question - even if the cards are played face up, is there always a clear, unambiguous way that you would play them? For example, playing 1-2 six-handed where everyone has stacks of 300, you are UTG with 9d-9c, the cutoff has Qh-Jh, button has Ks-10s, and the rest have offsuit diamond and club rags (so there are no burnt hearts, spades, A, K, Q, J, 10, or 9).
According to the poker odds calculator, your 9-9 is a 51.78% to 48.22% favorite heads up vs. Qh-Jh.
According to the poker odds calculator, your 9-9 is a 52.57% to 47.43% favorite heads up vs. Ks-10s.
However, against both opponents, your 9-9 has a 30.16% chance to win, Qh-Jh has a 34.95% chance to win, and Ks-10s has a 34.89% chance to win.
So what is your perfect play with 9-9? With Qh-Jh? With Ks-10s?
I'm imagining there are a lot of these relatively close situations with a small to medium pair vs. one or more opponents holding two overcards where, per the Sklansky theorem, one would be hard pressed to conclude the "winning" or "losing" play (fold, call, or bet/raise and exactly for how much?) for each even if everyone were playing with their cards up.