{Split from Fnord's winner take all thread} - ICM discussion

[StarTracker]

I don't use ICM because I haven't been convinced it is a winning strategy. Looks to me like it is more like an excuse when you lose.

Not sure I understand this, can you elaborate? Is it the specific model you don't like or the general principles?

I don't have a problem with aids being used, if that is what you mean by general principles. Actually I view the use of aids as a way to bring more money to the table. People are willing to risk more if they perceive they have an advantage.

I do have a problem with the ICM as I understand it. It seems to make some big assumptions, like putting someone on a range of hands. I think it is a major error to leave out skill. Also, a linear map of stack size to percent of prize pool does not take into account the power of the larger stack. It should probably be more based on simulation.

So, I think ICM has good intentions, but in the end it has too many flaws for me to use. If it was perfect and it told me exactly what to do in every situation, then it would take me out of the game and reduce the game to robotic pushing, and that is no fun. Since it isn't perfect, I kindof like people using it against me.

[Fnord]

StarTracker, what level are you beating?

[CoccoBill]

It seems to make some big assumptions, like putting someone on a range of hands. I think it is a major error to leave out skill.

2 questions:

- what do you put your opponents on then if not ranges?
- could you give me an example of this "skill"?

[fjuanl]

I'm kind of surprised Stars/FT hasnt added this, I think it would be pretty popular

As far as the ICM comments, I agree about its error in not taking skill into account, although its rare that someone is so grossly bad that it should affect your decision. For example, you have a marginal decision on the bubble and know that 2 other players don't take into account bubble factors and therefore are more likely to get in an all-in showdown. But I don't know if that and the fact that it might undervalue bigstacks means you should completely ignore it. I'd suggest posting a thread on 2+2 because they can probably come up with better arguments/ ICM is their god

[StarTracker]

Fnord, I am consistantly beating the $15 level. Moved up from the $1 where I started four years ago. Recently I moved to multi-table turbos and have increased my variance, but have found them softer.

CocoBill,

- what do you put your opponents on then if not ranges?

Like I said, this is based on my understanding of ICM. I could be totally wrong. I am assuming a realtime ICM calculator will determine a range of hands someone could push with, and will say what range of hands you can call with. I think SNGWiz will do this after the fact for training. I don't think it will allow you to put him on a range yourself and let it tell you your call range. However, if it is based on the range you put him on, then the ICM is very much dependent on your own skill in determining the range of hands, which is a good example of where skill can play a role. The more it calculates for you, the less skill involved.

I just think it (ICM) is an oversimplication of a very complex problem, and just because it is a complicated algorithm which does a lot of work to reduce it all down to a simple set of actions, doesn't mean the answer is correct. Anyone know of any studies done on the effectiveness of ICM?

Regarding 2+2, that forum is the reason I am here. I ratio of good info to bad there made it not worth the effort to read anything. Much better good to bad ratio here.

[FlyingSaucy]

ICM is still relevant with winner take all. As Fnord said originally, it becomes "less" relevant. ICM is an important factor for any game where placement pays and chips won are not directly equivalent to cash won.

I'd suggest posting a thread on 2+2 because they can probably come up with better arguments/ ICM is their god

What is wrong with FTR? They are right to put ICM in high regard, and most sng players here also do. So, can we not come up with the arguments ourselves or do we have to rely on douches to tell us?

[taipan168]

Startracker, I think you are getting a couple of things mixed up.

- ICM is just a mathematical model, just like algebra, for estimating the $ value of certain chip stacks in a tourney. Eg. if you have a 3000 stack on the bubble and the other stacks are 6000, 3000 and 1500 then your stack is worth 25.1% of the prize pool.

- ICM calculators like SNGPT or SNG Wiz are absolutely dependent on determining opps' calling or shoving range. Garbage in = garbage out. Whilst SNG Wiz does suggest a range, I almost always over-ride the default settings

As for the shortcomings of ICM, this is what Slim Pickens wrote some time ago on 2+2:

ICM has been demonstrated to be reasonably accurate, at least as benchmarked against some better (but much more computationally and theoretically intensive) models for most cases. It also models a fairly large sample of real-world SNG data well. There is still argument about some limiting cases and exactly where the model breaks down, but for pretty much any case that matters to the average players it does just fine.

[The shortcomings of ICM are:]
* The error in an ICM-calculated prize pool equity, relative to it's magnitude, gets large when the chip stack involved becomes much smaller than the big blind. This error isn't particularly important to the equity of the big stacks at the table, but it can be important in cases where there are multiple short stacks on the bubble.
* The ICM doesn't account for position, and any attempt to correct it is basically an empirical parameter fit. If position is somehow important in a situation, it won't be reflected in the ICM-calculated prize pool equity. A pair of 2 BB stacks on a 4-handed bubble will probably have different actual values depending on who is going to hit the blinds first.
* If there is a large skill differential, the ICM won't capture it. For example, there are times when a shorter stack can be played much more skillfully than a larger stack because a short stack and a rebuy ticket allows a skilled player to bluff/semi-bluff a lot more hands.

I would add to this the fact that the value of having a big stack on the bubble is not taken into account, but the incremental value is often not as much as most people think (because, for one thing, it depends on your opponents understanding that they can't make retarded calls of your pushes).

[StarTracker]

Thanks taipan, all this info is good stuff. I think I read that Slim article one time at 2+2 in the FAQ. I tried to add ICM to StarTracker as a way to help with decisions. I didn't feel right with it. First time I saw ICM was when they were chopping pots at the end of the big tourneys at 'Stars. Then it became a basis for bubble play. The resistance to the theory of bubble play is similar to the resistance it had in chopping pots.

Once the blinds and get high enough, ICM calcs to determine push, call, fold gradually becomes less and less relevant as the blinds increase. In a 150 hand game...typical ICM calcs are useful in a small period of time around the bubble.

[taipan168]

Once the blinds and get high enough, ICM calcs to determine push, call, fold gradually becomes less and less relevant as the blinds increase. In a 150 hand game...typical ICM calcs are useful in a small period of time around the bubble.

I disagree with this - ICM is ALWAYS relevant, particularly in 1-table SNGs - just that the degree of divergence between chip EV and $ EV is greatest on the bubble and particularly when there is a short stack around. Also, ICM calculations are still very important when blinds are large relative to stacks and you are making push/fold decisions - in fact this is where a program like SNG Wiz is most useful.

A couple of examples as to why ICM important at all stages of the game.

1. First hand of the game, you're in the BB with 33 and it's folded to the SB who shows you AKo and shoves all-in. Do you call?

2. It's the bubble and UTG has gone berserk with his shoving, having shoved the last 10 hands in a row. Chip distribution is: UTG 6000, Button 2000, SB 1500 and you in the BB with 4000 and AKo. Blinds are 200/400. True to form, UTG shoves again. Do you call?

(Answers after we get some replies).

Also, I am not sure how you would incorporate ICM into StarTracker, since StarTracker is a results tracking website?

[Fnord]

1. First hand of the game, you're in the BB with 33 and it's folded to the SB who shows you AKo and shoves all-in. Do you call?

2. It's the bubble and UTG has gone berserk with his shoving, having shoved the last 10 hands in a row. Chip distribution is: UTG 6000, Button 2000, SB 1500 and you in the BB with 4000 and AKo. Blinds are 200/400. True to form, UTG shoves again. Do you call?

I assume both are a fold. This is the crux why I dislike the format for online play. Any online game that rewards folding this much is pretty much fucked because it's sooo easy to fold while playing 8+ tables. However, live is a different story because so few people are capable of making good folds for over an hour then on top of that switch gears to make the correct shoves.

[CoccoBill]

Like I said, this is based on my understanding of ICM. I could be totally wrong. I am assuming a realtime ICM calculator will determine a range of hands someone could push with, and will say what range of hands you can call with. I think SNGWiz will do this after the fact for training. I don't think it will allow you to put him on a range yourself and let it tell you your call range. However, if it is based on the range you put him on, then the ICM is very much dependent on your own skill in determining the range of hands, which is a good example of where skill can play a role. The more it calculates for you, the less skill involved.

Well, you are totally wrong. Forget SNGWiz and SNGPT when talking about ICM, they're not the same thing. ICM is a mathematical model that determines the value of your chip stack in relation to other stacks in the table, nothing more. You can, however, use it to help in difficult decisions, since it allows you to calculate what your chip stack in relation to others will be, whether you fold/bet/raise/win/lose a hand. In effect it tells you whether the risk is worth the reward.

SNGWiz/SNGPT then take this model, allow you to assign ranges on opponents and "pokerstove" the results weighed by the ICM model, and give you vital information on making the correct decisions. As taipan already said, if you're not able to assign proper ranges for the opponents, the results will also be wrong.

[StarTracker]

Well, I'm glad I am totally wrong.

While I was looking into ICM, I found two links which provided the best info:

The Chillin411 site, which has some good hand vs. hand tables which really can reduce the 'pokerstove' calc time.
Some Java code for the actual implementation of ICM.

Seeing the actual code was the only way I could understand what it was doing.

Yes, StarTracker is a results database. I have a program in beta test which shows a summary of hands played for each player at your table, along with their STR number. I have access to the hand history for all hands played at the table, and was going to incorporate some ICM advice based on stack sizes and position I know about.

Separating the ICM calculation from the push/call/fold advice, I still am not convinced ICM is a good estimate of equity. It doesn't seem to be a linear problem as I read it. For example, once the game gets headsup, difference in skill level becomes more of a factor. How many times have you gotten HU with a donk who lucked his way into a huge chip lead and taken it down because he was an idiot? Sometimes it is good to let the luck sack build a good chip lead and knock out good players. Then when you are HU, the chips are easier to obtain. Or when Eric Seidel prolonged the bubble by checking the nuts on the river so the bubble would stay alive and he could continue to command the table.

Stack sizes do have something to do with it, but I have seen more than enough times the order of stacks on the bubble do not map to the eventual order of finish. I know that is an oversimplification of ICM, but bare with me.

Where a player ultimately finishes in a tournament is determined by the following ranked in decreasing influence:

1) His skill.
2) Skill of other players.
3) Where skilled/unskilled players are relative to him.
4) His cards (actually an equalizer since everyone is in the same situation.)
5) Where the big stack is relative to him. On his left is negative, on his right is positive.
6) Where the loose/tight players are relative to him. Loose on right, tight on left is positive. Opposite is negative.
7) His stack size relative to others.

So, ICM is estimating my equity based on what I rank as having the least impact on determining the actual finish position. It just seems to reduce the whole thing down to a matter of cards (equal) and chip stack sizes. Yes, you can add some skill to this by accurately putting people on range of hands and how likely someone is to lay it down. This knowledge is then plugged into a calculation which seems to always come up with a borderline decision. Maybe someone can show some examples where the decision is very clear, but based in the way equity is calculated, it seems like it would muddle it all down to a small increase/decrease in equity.

1. First hand of the game, you're in the BB with 33 and it's folded to the SB who shows you AKo and shoves all-in. Do you call?

2. It's the bubble and UTG has gone berserk with his shoving, having shoved the last 10 hands in a row. Chip distribution is: UTG 6000, Button 2000, SB 1500 and you in the BB with 4000 and AKo. Blinds are 200/400. True to form, UTG shoves again. Do you call?

1. I know what you are getting at, but it is hypothetical in that the guy has shown his cards.

2. I am going to guess that a call is in order. What other hand are you going to call with? What does something like SNGWiz say is the appropriate action. Can you show the calcs?

[CoccoBill]

I think you're still confusing things.

"ICM computes the tournament equity that a certain chip stack has in relation to the other stacks at the table in order to determine the true value of the stacks."

That's ICM, nothing more, nothing less. Everything else you're trying to describe here are specific situation-based calculations implementing ICM. "ICM calculations" are applicable (and meant) for a specific SNG situation, when stacks in relation to blinds are getting short. They're not meant for early game, and not for HU. While the push/fold calculations are not valid throughout the whole tournament, the ICM model itself still applies from the first hand of an MTT to the last hand of HU.

[StarTracker]

"ICM computes the tournament equity that a certain chip stack has in relation to the other stacks at the table in order to determine the true value of the stacks."

I understand that is what ICM does and I have actually implemented the JAVA code above in C#. So, I do know what it is.

I still do not think it is a good way 'to determine the true value of stacks.' See my reasons above.

I also understand that you can use ICM to determine your change in equity due to what-if scenarios, and the probability of those scenarios. I guess the only name we have for this decision making process is SNGWIZ since that program seems to do a good job of doing all the math.

The probabilities of people folding, of your hand winning or losing is all basic probability calculations, and I can't argue with that. The problem I have is that the ultimate decision is based on a flawed model of determining the true value of stacks.

[CoccoBill]

Certainly ICM has its limitations and even some flaws, since it simply cannot factor in all the variables, noone is denying that. If you have a better mathematical representation in mind, please tell us.

Even with its flaws its still the best method for estimating the equity we have in SNGs. The shortcomings can partly be remedied by using the Minimum Edge setting (in SNGPT, don't know what its called in SNGWiz), that is, if you're playing against better players than you, you might want to take every even slightly +EV chance, but if you have the skill advantage you can lay down the marginal ones.

[taipan168]

Good discussion! I'll address the worked example first and the general discussion later.

1. First hand of the game, you're in the BB with 33 and it's folded to the SB who shows you AKo and shoves all-in. Do you call?

2. It's the bubble and UTG has gone berserk with his shoving, having shoved the last 10 hands in a row. Chip distribution is: UTG 6000, Button 2000, SB 1500 and you in the BB with 4000 and AKo. Blinds are 200/400. True to form, UTG shoves again. Do you call?

1. I know what you are getting at, but it is hypothetical in that the guy has shown his cards.

Humour me for a bit. According to Pokerstove, 33 is 53.4% to win against AKo. If we were playing a cash game, we would call here because we're getting even money when we're better than even money to win this hand. So, do we call in a tourney?

2. I am going to guess that a call is in order. What other hand are you going to call with? What does something like SNGWiz say is the appropriate action. Can you show the calcs?

- If we fold we will have 3600 chips
- If we call and win we will have 8200 chips
- If we call and lose we are out.

Therefore, given that opp is probably shoving any two cards, in a cash game we would instacall because we are getting 1.27 to 1 on the call when we are 65.3% to win the hand. HOWEVER, in a tournament, incremental chips you win are worth less than incremental chips you lose (which is the key outcome of ICM) so we need to restate chip stacks in terms of % of the prize pool:

- If we fold we will have 3600 chips worth 28.6% of the prize pool
- If we call and win we will have 8200 chips worth 40.6% of the prize pool
- If we call and lose we are out.

Therefore, we actually need to be (28.6/40.6) = 70.4% to win to make this a good call. Against a random hand, AKo is only 65.3% to win, so perverse as it sounds, we need to fold our AKo. Our call range in this spot is 99+.

[taipan168]

Where a player ultimately finishes in a tournament is determined by the following ranked in decreasing influence:

1) His skill.
2) Skill of other players.
3) Where skilled/unskilled players are relative to him.
4) His cards (actually an equalizer since everyone is in the same situation.)
5) Where the big stack is relative to him. On his left is negative, on his right is positive.
6) Where the loose/tight players are relative to him. Loose on right, tight on left is positive. Opposite is negative.
7) His stack size relative to others.

Let's take another example. It's down to four players in a 1-table SNG. Blinds are 200/400. The stacks, reads, skill levels (I'll use your very good STR numbers) and order of the players are:
1. Huge calling station donkey (STR 5) who has luckboxed his way to a 6000 stack
2. LAGGtard (STR 25) who had the big stack but got sucked out on by the donk: 3000 stack
3. Myself (STR 98): I've managed to shove a few hands to keep myself at my starting stack (1500) but just haven't picked up any cards and have missed some pushing spots because the donk has limped in in front or the LAGGtard has shoved.
4. Weak/tight player (STR 75) who more than doubled up early but has let himself get blinded down since then: 3000 stack

Can we really say that I have the best chance of winning this tournament because my skill level is way above all the other players? I am desperately short and need to find a spot to double up to just stay alive very soon and I'm unlikely to get a chance to shove into the weak/tight player because of the loose players in front of me.

So, ICM is estimating my equity based on what I rank as having the least impact on determining the actual finish position. It just seems to reduce the whole thing down to a matter of cards (equal) and chip stack sizes. Yes, you can add some skill to this by accurately putting people on range of hands and how likely someone is to lay it down.

The other way you incorporate skill into ICM calculations is, as CoccoBill said, to not take very marginal edges when you believe that you are likely to have less marginal edges later in the game.

This knowledge is then plugged into a calculation which seems to always come up with a borderline decision. Maybe someone can show some examples where the decision is very clear, but based in the way equity is calculated, it seems like it would muddle it all down to a small increase/decrease in equity.

SNGs are games of very slim edges, small increases/decreases in equity is all you can realistically expect.

As CoccoBill said, ICM is not perfect but it is the best compromise between simplicity and accuracy (there are other slightly more accurate models but they are hugely more computationally complex). Certainly it is far superior than just using plain chip EV.

[StarTracker]

You guys make some good arguments for the effectiveness of using ICM in making decisions at the table. I can see how it is better than chip EV but it is still not something which you can do in your head in the small amount of time you have to make your decision.

Does running something like SNGWiz after the fact help you make these decisions? Or is it still something which requires the ICM and the pokerstove calculations in real time. I have read a few books, and when it gets down to the detail of determining the EV of a move, I am always saying to myself 'no one does that in their head'. At least not with that much accuracy. But it requires accuracy because of the small edges mentioned.

[CoccoBill]

Does running something like SNGWiz after the fact help you make these decisions? Or is it still something which requires the ICM and the pokerstove calculations in real time. I have read a few books, and when it gets down to the detail of determining the EV of a move, I am always saying to myself 'no one does that in their head'. At least not with that much accuracy. But it requires accuracy because of the small edges mentioned.

After reviewing a few hundred tourneys you'll start to get a hang of it. I can whole-heartedly recommend getting SNGWiz or SNGPT (I'm using SNGPT myself), download the free trial of SNGWiz and play around with it, adjust opponents ranges, blind levels and position and see how it affects the results. Knowing proper push/fold game is essential to SNGs, even if you're not going to follow it every hand it's vital to understand it IMO.

[taipan168]

I don't use an ICM calculator whilst playing, I use one after my tourneys have finished to review hands which I think are marginal. In any case, Stars has banned the use of ICM calculators whilst the Stars client is open (which I think is a good development as it makes it much harder to run a bot).

The aim of reviewing hands is to develop your "feel" for +EV and -EV situations. Once you review a few hands you will be able to formulate general "rules" that help guide your decision making during play, such as:

- Don't call an AI on the bubble with an unpaired hand if you're the second stack, there are 1+ short stacks around and you're covered by the shover
- If you're UTG on a full or near-full table and you have ~10x BB left, your shoving range should be something like 77+, AJ+
- If it's folded to you in the SB and either you or the BB has 10x BB or less in your stack, it is generally +EV to shove ATC unless they're very loose
- If it's folded to you on the button and the SB and BB have roughly equal stacks to yours, it is generally +EV to shove ATC if you have 3-5x BB

The other thing that SNG Wiz allows you to do to develop this intuition is to change the variables of the hand to see what it does to the EV of the situation:

- If my opps call looser/tighter
- If my stack was shorter/larger
- If the hand is a push, what range could I shove here before it becomes a fold (and vice versa)
- If my position on the hand was different, what impact does that have?
- If the stack sizes were different, etc.

[StarTracker]

I think I am starting to see the value ( ) here.

One thing I have not been able to understand is how the ICM Calculator part assigns a probability that my opponent will call. Does it use his stack size, looseness/tightness or combination. Then, the percentage of times he does call, how do it assign a range of hands to 'pokerstove' it against your hand?

[CoccoBill]

I think I am starting to see the value ( ) here.

One thing I have not been able to understand is how the ICM Calculator part assigns a probability that my opponent will call. Does it use his stack size, looseness/tightness or combination. Then, the percentage of times he does call, how do it assign a range of hands to 'pokerstove' it against your hand?

You still got it a bit upside down. You yourself assign your opponent the calling/shoving range, and it does the "pokerstoving" against that. It doesn't itself figure out or guess anything, purely does the calculations based on the data you provide.

[StarTracker]

You still got it a bit upside down. You yourself assign your opponent the calling/shoving range, and it does the "pokerstoving" against that. It doesn't itself figure out or guess anything, purely does the calculations based on the data you provide.

I went ahead and got the 30 Day free trial of SNG Wiz. I found out they do opponent modeling as an initial first guess, but strongly recommend that you put opponents on ranges of hands yourself. They do use some criteria in their opponent modeling to put people on a range of hands.

What criteria do you use to put people on hands?

They also are gathering Hand History statistics (no individual profiling, just general stats on call/push ranges). Which seems very interesting.

Still trying to figure out how to run it, it is not very obvious to me. Guess I will have to dig through the docs.

[CoccoBill]

What criteria do you use to put people on hands?

I'd say its a combination of (in no particular order) the "typical" ranges, opponent's stats, prior action, stack sizes, stack/blind ratios (M if you wish), number of opponents left, position, payout structure, reads (e.g. does opp like to play low aces/SCs/etc) and gut feeling. I don't know how SNGWiz does the modeling, actually I wasn't aware it does it at all. I use SNGPT and it doesn't, it just has some "default" ranges you can assign, but I don't use them much. Whether my self assigned ranges are closer to the truth than the default ones might be debatable, but at least I'm doing the dirty work myself and possibly even learning something in the process.

[StarTracker]

Hi, I spent a big part of yesterday reviewing my mistakes and then played 3 sit&gos last night. I got 1st in a 1 table, 2nd in a 180 and 1st in a 2 table. So, a big thank you to SNGWiz! Since I lasted so long, I have a ton of mistakes to review today.

I wrote down one hand while playing which I wanted to review later. I had 99 in mid position, less than 10x BB, and two stacks with almost double my stack in the blinds. I folded, but SNGWiz said I should push here with 22+. I also thought with 4 players yet to act, SNGWiz would tell me to fold. But no.



I also had the SB as very loose and bad, I have been wanting to get his chips.

Real cool program they have. I am very impressed now that I figured out how to use it.

[fjuanl]

SNGWIZ was mainly created for 1-table SnG's. MTT's have a more top-heavy payout structure so you can take smaller edges and call/shove wider near the bubble

[taipan168]

Yeah, you need to make sure that you have the correct payout structure set in SNG Wiz - it's really important when doing the ICM calculation. Not sure how to do it but the SNG Wiz site explains it.

Glad that you're finding SNG Wiz and ICM to be useful tools!

[StarTracker]

I was verifying that when I saw the advice. I submitted a bug to SNGWiz which they are looking into. This was before the final table, however, you will notice I have the setting on Chip Equity, and not $. And, looking at their excellent breakdown in scenarios, it shows how I have a 43% chance of taking it right then, and I have right around a 60% chance of winning if some loose guy looks me up.

I see this as a good example of how SNGWiz can change your attitude about medium pair against big stacks.

[Outlaw]

I agree with you in a lot of ways. The biggest problem I find in ICM is it will not help a person learning the game assign ranges to people. Figuring out someones range depends so much on the flow of the game and table image. I feel it might hinder a new player to learn ICM too early as it will make the game too black and white. Poker is not played by computers, its played by people.

I had never heard of ICM until I signed up for FTR after having played for 7-8 months. I downloaded SNGET and have spent a lot of time since practicing with the SNG Quizzer than comes with the program. I found that the first time I used it I was able to consistently get the ICM questions right 85-90% of the time just using my experience as a guide. I have no idea about the math behind ICM, I flunked math all the way through school. All I know is that most of the time it just makes sense on what to call or push with based on your hand, the stack sizes, etc.

All that being said, I sometimes am puzzled at some of the supposedly mathematically "correct" plays according to ICM. Sometimes they just appear to be blatently wrong. Maybe its because ICM can't take into account minute details other than an imprecise range of cards? (Imprecise as you can never truly know someone's range and have to rely on other factors.)

For example: I was using the quizzer today and it was basically this setup. 3-handed, I was the BB with AJo, blinds 100 200. The button had pushed with a range of the top 16% of hands (33+, A7o+, A3s+, KQo, K10s) I had a stack of 3400, the SB had 6700, and the button 9600. Supposedly, the correct play was to fold! There is no way in heck I am folding here as the short stack. Even though the blinds are not yet huge, they will be soon. I am not going to lay down a top 7% hand at this moment. Even though I may have the button's range pegged at 16%, we all know as a big stack you tend to play a little looser than your image. And this is where it gets complicated and ICM falls short, in my opinion. ICM says its -ev, I say its +ev.

Back to ranges, that is the entire X factor here. Its really impossible to know someone's range as everyone has differing preferences for cards. Some might not push with KJs but they would push with 56s, for obvious reasons. Yet another person might do the exact opposite. If you tried to assign them a range to stick into an ICM calculator it'd be very difficult.

I am not arguing that doing the mathematically perfect +ev play every hand will not win money in the long run.. it will. I am just saying that one can never know all the info and sometimes its better to just go with your "gut" rather than what the robot tells you to do.

How do we achieve this "gut" instinct? By playing millions of hands and thinking about each and every one of them. Doing it so many times and recording the results in the back of your head time and time again. Eventually a knack develops for being able to absorb all factors at once and instantly knowing the correct play. At some point you may seem to act as a robot, but its only because the intuition you've developed let's you know the correct play.

[taipan168]

All that being said, I sometimes am puzzled at some of the supposedly mathematically "correct" plays according to ICM. Sometimes they just appear to be blatently wrong. Maybe its because ICM can't take into account minute details other than an imprecise range of cards? (Imprecise as you can never truly know someone's range and have to rely on other factors.)

The classic situation where people think that ICM is "wrong" is the calling AI with AK situation that I described above.

Here was a long discussion about a similar situation.

For example: I was using the quizzer today and it was basically this setup. 3-handed, I was the BB with AJo, blinds 100 200. The button had pushed with a range of the top 16% of hands (33+, A7o+, A3s+, KQo, K10s) I had a stack of 3400, the SB had 6700, and the button 9600. Supposedly, the correct play was to fold! There is no way in heck I am folding here as the short stack. Even though the blinds are not yet huge, they will be soon. I am not going to lay down a top 7% hand at this moment. Even though I may have the button's range pegged at 16%, we all know as a big stack you tend to play a little looser than your image. And this is where it gets complicated and ICM falls short, in my opinion. ICM says its -ev, I say its +ev.

The answer to your question is right here - the range you are using for the big stack is too tight.

[CoccoBill]

Maybe its because ICM can't take into account minute details other than an imprecise range of cards? (Imprecise as you can never truly know someone's range and have to rely on other factors.)

Absolutely, but the closer you get, the better your results will be (whether you use ICM or not).

For example: I was using the quizzer today and it was basically this setup. 3-handed, I was the BB with AJo, blinds 100 200. The button had pushed with a range of the top 16% of hands (33+, A7o+, A3s+, KQo, K10s) I had a stack of 3400, the SB had 6700, and the button 9600. Supposedly, the correct play was to fold! There is no way in heck I am folding here as the short stack. Even though the blinds are not yet huge, they will be soon. I am not going to lay down a top 7% hand at this moment. Even though I may have the button's range pegged at 16%, we all know as a big stack you tend to play a little looser than your image. And this is where it gets complicated and ICM falls short, in my opinion. ICM says its -ev, I say its +ev.

You're 51.4% to win against his range and even if you double up you're by no means guaranteed 2nd place, you have more to lose than to gain. If BTN does in fact shove 23%+ here it would make the call correct, but based on that data it is a fold. If you punch wrong numbers in a calculator there's a pretty good chance it won't give the correct result.

Back to ranges, that is the entire X factor here. Its really impossible to know someone's range as everyone has differing preferences for cards. Some might not push with KJs but they would push with 56s, for obvious reasons. Yet another person might do the exact opposite. If you tried to assign them a range to stick into an ICM calculator it'd be very difficult.

Yes, that's where the poker "skill" is supposed to come in.

I am not arguing that doing the mathematically perfect +ev play every hand will not win money in the long run.. it will. I am just saying that one can never know all the info and sometimes its better to just go with your "gut" rather than what the robot tells you to do.

Why would your gut be any more correct about the whole "calculation" than about assigning a simple range? Anyway, you're not supposed to use ICM calculations during play (in fact that's against the policy of most sites), but to review your play afterwards and school your gut.

How do we achieve this "gut" instinct? By playing millions of hands and thinking about each and every one of them. Doing it so many times and recording the results in the back of your head time and time again. Eventually a knack develops for being able to absorb all factors at once and instantly knowing the correct play. At some point you may seem to act as a robot, but its only because the intuition you've developed let's you know the correct play.

Intuition, by definition, has no objective validity. It dates back to the stone age, where reaching a decision (any decision) quickly was paramount for survival. A bear is attacking and you have a club in your hand, do you hit it or run for it? Do neither and die for sure. Colbert might disagree with me, but yes, the gut can be wrong. Actually, the less understanding of the related concepts you have, the more likely it is.

[Outlaw]

Why would your gut be any more correct about the whole "calculation" than about assigning a simple range? Anyway, you're not supposed to use ICM calculations during play (in fact that's against the policy of most sites), but to review your play afterwards and school your gut..

All I was saying is that you learn adjust your opponents range based on reads from thousands of past situations... just assigning an arbitrary range to an opponents based on their history of whether they are loose or tight is inferior.

Also, you can use the calculations and long as you figure them yourself and not use a computer program to do it. Actually Pokerstars allowed you to use an ICM program up until the 1st of this month.

Intuition, by definition, has no objective validity. It dates back to the stone age, where reaching a decision (any decision) quickly was paramount for survival. A bear is attacking and you have a club in your hand, do you hit it or run for it? Do neither and die for sure. Colbert might disagree with me, but yes, the gut can be wrong. Actually, the less understanding of the related concepts you have, the more likely it is..

I agree with this point. A player that understands all of the pieces of the puzzle can make a much better "gut" decision. Someone ignorant to the relevant concepts is taking a blind stab in the dark whether using so-called intuition or not. He might as well give a monkey a stick and let him choose.

[StarTracker]

Here is a good paper which actually tests ICM out in a simulation of play.

Computing an Approximate Jam/Fold Equilibrium for 3-player No-Limit Texas Hold’em Tournaments

The conclusions were interesting: Traditional hand rankings didn't hold up, and ICM Equity calculation broke down in some cases.

[StarTracker]

I am bumping this thread because it has a lot of background for what I am about to present.

Back at the beginning of the thread, one of my objections to ICM is the way it determines equity based on stack sizes. ICM is a way of using stack sizes to predict how often a player will finish in each position and determine equity from those probabilities.

For example: If there are 4 players left in a sit&go, and they all have the same size stack, they all have a 25% equity in the prize pool. If the stacks were different, then you might get this:



The player with 6000 chips as 36% equity according to ICM. ICM basically assigns a probability of finish position based on stack size. Stack size does have something to do with it, however, this just seemed to arbitrary to me. There has got to be more to it. As someone said, well it is as good a model as we can get.

I think I have come up with a better way of calculating equity. I call it Real Equity and you can see it in the small spreadsheet I posted. This Real Equity is calculated by simulating 2000 sit&gos starting with the current stack sizes at a certain blind level. It uses basic push/fold/call logic based on hand values to play out the rest of the sit&go.

In the example above it gives more equity (17%) to the small stack vs. ICM (6%) and less equity to the big stack (29%) vs. ICM (36%).

This is a pretty good example of the difference between Real Equity and ICM Equity. So, here are some questions:

1) How does using Real Equity change the calculations of whether or not you should push or fold? Do small stacks tighten up or do they loosen?

2) Anyone have any ideas on improving the simulation? Is there a basic game play strategy which I can use to get a better simulation?

3) And a chicken or the egg thought...if you base your push/fold decisions on equity numbers calculated from normal play, then isn't your equity underestimated in that model. You use the model to improve your equity. So, should you have a different push/fold strategy for you and normal for the others? Or, should you just plain have an ISM (Independent Strategy Model) for the simulation?

I like the simulation approach to determining equity as opposed to the recursive stack size probability calc in ICM. Seeing the big differences in equity was encouraging. Perhaps there really is a better way.

[vqc]

I am bumping this thread because it has a lot of background for what I am about to present.

Back at the beginning of the thread, one of my objections to ICM is the way it determines equity based on stack sizes. ICM is a way of using stack sizes to predict how often a player will finish in each position and determine equity from those probabilities.

For example: If there are 4 players left in a sit&go, and they all have the same size stack, they all have a 25% equity in the prize pool. If the stacks were different, then you might get this:



The player with 6000 chips as 36% equity according to ICM. ICM basically assigns a probability of finish position based on stack size. Stack size does have something to do with it, however, this just seemed to arbitrary to me. There has got to be more to it. As someone said, well it is as good a model as we can get.

I think I have come up with a better way of calculating equity. I call it Real Equity and you can see it in the small spreadsheet I posted. This Real Equity is calculated by simulating 2000 sit&gos starting with the current stack sizes at a certain blind level. It uses basic push/fold/call logic based on hand values to play out the rest of the sit&go.

In the example above it gives more equity (17%) to the small stack vs. ICM (6%) and less equity to the big stack (29%) vs. ICM (36%).

This is a pretty good example of the difference between Real Equity and ICM Equity. So, here are some questions:

1) How does using Real Equity change the calculations of whether or not you should push or fold? Do small stacks tighten up or do they loosen?

The fact that "real equity" and ICM equity have different values does not necessitate changing the METHOD in which you calculate what your move should be. They are just equity models.

Thus, whether small stacks should tighten or loosen up does not depend on HOW the equity is being calculated, but rather WHAT their equity is going to be (which depends on the different outcomes).

[StarTracker]

Thus, whether small stacks should tighten or loosen up does not depend on HOW the equity is being calculated, but rather WHAT their equity is going to be (which depends on the different outcomes).

The different outcomes of the specific hand in question? Decisions on the hand in question use EV to make a decision on proper action. We do that which will give us the greatest EV. Equity is a part of the EV calculation along with probabilities of other folding/calling and how the board can play out if it goes to show down. I am just talking about the equity part of the EV calculation. ICM is HOW traditional EV calculators (ie SNGWIZ) determine equity.

[taipan168]

StarTracker, something looks weird to me about the 600 stack having 17% equity under your simulation compared to 6% equity under ICM. I just cannot imagine that the 600 stack will finish in the money often enough to make his/her stack worth 17% of the prize pool. Perhaps you are assuming that the short stack pushes very frequently but the other stacks call very tight?

Some of this will obviously depend on how big the blinds are because if they are 10/20 the 600 stack obviously has much more time to wait for a hand than if they are 200/400.

[StarTracker]

Taipan168, you might be onto something. I should take into account the short stack push range and the odds when deciding whether a big stack should call.

In the meantime, do you think this is a better way to determine equity?
I think it will get down to position causing a difference in models.
Maybe I get something close to ICM anyway, at least I tried.

[vqc]

Thus, whether small stacks should tighten or loosen up does not depend on HOW the equity is being calculated, but rather WHAT their equity is going to be (which depends on the different outcomes).

The different outcomes of the specific hand in question? Decisions on the hand in question use EV to make a decision on proper action. We do that which will give us the greatest EV. Equity is a part of the EV calculation along with probabilities of other folding/calling and how the board can play out if it goes to show down. I am just talking about the equity part of the EV calculation. ICM is HOW traditional EV calculators (ie SNGWIZ) determine equity.

Cliff notes: I think its a semantics problem.

The method you use to calculate equity ("real" vs. icm) has no effect on the expected value, unless the results of the methods are different.

Thus, a question of "HOW does using X equity calculation method change whether or not you should push or fold" doesn't make sense to me. I think it would make more sense if you had asked "Given that Real Equity tends to INCREASE the equity of smaller stacks relative to the equity projected by ICM, how does this affect the play of smaller stacks." Even then the answer would be dependent on specific situations.

[StarTracker]

This thread is filled with semantic confusion.

In any event, you restated my question perfectly:

"Given that Real Equity tends to INCREASE the equity of smaller stacks relative to the equity projected by ICM, how does this affect the play of smaller stacks?"

The method you use to calculate equity ("real" vs. icm) has no effect on the expected value, unless the results of the methods are different.

Yes, if they return the same numbers, no difference. After tweaking my simulator (allowing the big stack to call when he has odds), I still see some differences between Real and ICM Equity calcs. I will post more about the new results this weekend.

The first time I heard of ICM, was when Lee Jones was chopping final tables at 'Stars a few years ago. He would come up with a chop proposal by typing the stacks into a spreadsheet, and the equity numbers would come out. ICM does bring the big stacks down, and the small stacks up (as compared with basic chip percentage) which is expected, however, I feel position of your stack to the small stack or big stack has an impact on your equity also.