ICM/SNGPT rambling thoughts(long)

[ QUOTE ]
[ QUOTE ]
[ QUOTE ]
Raptor is right that playing a certain hand right culminates in getting right the calling ranges of your opponents. But he is only almost there. You have to factor in what your decision will mean to the future calling ranges and how this will affect your EV on your future weighted average decisions. If opponents loosen up, your average EV will very often go down which means that if you stay with the threshhold you will have less situations where you can push for profit.

[/ QUOTE ]

Excellent, excellent summation and something that is very difficult to model, but would be an excellent addition to the current state-of-the-art ICM model (which includes more than just ICM, but all the extra stuff that eastbay has packed into SnGPT) if we could come up with something reasonable.

[/ QUOTE ]

This may be the best point to come from this discussion. It's not where my brain was headed when I started but I like this idea.

Key factors in an ICM calculation are the size of the blinds
and your FE vs. villian's range. Large +$EV situations almost always involve large blinds. Most pushes involve the same 1 or 2 villians on your left.

So, you've pushed from SB into BB the last two orbits. Here you are again with a marginal(+.6%) push against his current range with 100-200 blinds. If you have a workable stacksize and you feel that pushing this hand will open villian's range drastically on future hands, can you argue that the lost FE in the future higher blinds out weighs the current slightly positive situation?

[/ QUOTE ]

Yes, but again, the effect is very small.

[Nicholasp27]

it predicts the future based solely on EQUAL SKILL LEVELS and EQUIVALENT TABLE-IMAGE...

do u agree that if u sit down next to 8 donks and gigabet that icm is correct in saying that everybody's ev is .1?

if not, then skill and table-image matter

and skill is not static...if an action during the time of the tourney (on this table, another, or non-poker related) occurs to change someone's actions (tilt, etc) then skill levels change...another thing that changes is table-image...if u push 27o and get caught, even if u win the hand, your future ev has gone down


do you agree that if i push 27o and get caught...then the ev of my pushing kjo next hand is less than if i folded that 27o? if so, then u agree that table image matters

ICM is a static predictor of the future, not dynamic...and it makes 2 HUGE assumptions: skill level and table image are equal

yes, no matter ur skill level, it helps to get an extra .1 ev...but if it hurts your table image, then it may not be worth the .1



again: i realize icm predicts the future, but only based on chip stacks and blinds...not based on skill/table image...both of which can change dynamically from hand to hand

[AliasMrJones]

[ QUOTE ]
So, you've pushed from SB into BB the last two orbits. Here you are again with a marginal(+.6%) push against his current range with 100-200 blinds. If you have a workable stacksize and you feel that pushing this hand will open villian's range drastically on future hands, can you argue that the lost FE in the future higher blinds out weighs the current slightly positive situation?

[/ QUOTE ]

I agree this is a very interesting topic and is very difficult to quantify. It is, unfortunately, not just the last few hands either. If someone has pushed UTG on my BB the last 3 times in a row (not 3 sequential hands in a row) that also tends to widen my calling range, for me quite a bit. ;-) To make it simple, perhaps there is a way to take previous X hands actions combined with current push = some widening of future calling range. I think, for the most part, the tighter our opponents are, the better as we can steal with a wider range of hands. So wider future calling ranges probably = smaller future +$EV.

Seems like it would be too hard to factor in much stuff manually so if the proggy were reading a hand history it could track things, but if doing it by hand, maybe a single table image number? The program could evaluate the impact of a push by assuming your image number goes down by some margin with every push and so tries to calculate the future $EV impact of this change? If you have a high table image number, impact isn't so big, but if your table image number is low, the change is greater. (Like if I push once there isn't much change (say 1 point), but each additional time a push is a large and larger change (like 2 points for the next push and 4 point for the one after that.) I think we're pretty much into wild ass guess territory at this point, though.

[Nicholasp27]

u'd also have a table image number of urself for each opp...if u've stolen bb's blinds the last 2 circuits, your table image for him is much smaller than for the other 2-3 players...

this will affect your future ev somewhat...how much? we don't know...but imo it's more than .1% in many instances

[BadMongo]

I don't understand why you are getting so hung up on table image... ICM accounts for table image through calling ranges.

When we use ICM to calculate the EV of a push/fold decision, there are only 2 actions our opponent can take against us if we push: 1) he can fold, or 2) he can call. Thus, our table image can ONLY affect our opponents' decisions by swaying them either towards a fold or towards a call. If it's the former, we tighten the opponent's range, and if it's the latter, we loosen it. So our table image has been incorporated into the model.

[AliasMrJones]

[ QUOTE ]
it predicts the future based solely on EQUAL SKILL LEVELS and EQUIVALENT TABLE-IMAGE...

do u agree that if u sit down next to 8 donks and gigabet that icm is correct in saying that everybody's ev is .1?

if not, then skill and table-image matter

and skill is not static...

again: i realize icm predicts the future, but only based on chip stacks and blinds...not based on skill/table image...both of which can change dynamically from hand to hand

[/ QUOTE ]

Well, it's really hard to discuss this with you because you keep changing your argument. You started off by saying it would be correct to pass up a +$EV opportunity for a greater +$EV opportunity later, which is clearly incorrect. Now, you are talking about factoring skill level and future table image into ICM.

Trying to factor skill level into ICM has already been discussed and agreed would be a nifty thing. I think trying to factor in impact on future table image would also be interesting.

However, the fact remains that a SnG is a very short period of time. How much impact your table image could possibly take during the period when ICM push/fold kicks in and the end of the tournament is probably not that great compared to the larger share of prize money you get from pushing when ICM says to push. (So, basically, I think there is a larger chance that you will overcompensate and lose prize money equity by folding when you really should be pushing than gain prize money equity by keeping your image up.)

These things are refinements and could have a small, but noticeable affect on ICM predictions. Given the short nature of push/fold play in a single SnG, I doubt the impact will be very large, however. Again, I would think less than 1 potentially changed decision per tournament on average.

[Nicholasp27]

ok, here is my understanding about ICM and SNGPT:

tell me which assumptions/statements are wrong


1) ICM takes the input of stack sizes of all players and the payout structure and outputs the expected prize equity that each stack has (predicts future)

2) ICM does not take into account blinds/skill/table image or anything other than stack sizes and payout structure

3) SNGPT uses ICM

4) SNGPT figures out the probability of occurences W, X, Y, Z: everyone folds to your push, people call you and you win, people call you and you tie, people call you and you fold

5) After doing #4, SNGPT plugs those resulting stack sizes into ICM to determine the equity in each of those scenarios

6) SNGPT then examines if ICM says your equity is higher if you fold or if you push based on probability of W/X/Y/Z occurring and their resulting ICM equity results

[AliasMrJones]

[ QUOTE ]
u'd also have a table image number of urself for each opp...if u've stolen bb's blinds the last 2 circuits, your table image for him is much smaller than for the other 2-3 players...

this will affect your future ev somewhat...how much? we don't know...but imo it's more than .1% in many instances

[/ QUOTE ]

No -- You don't care about your opponents table image. It has absolutely no impact on your prize money equity. Your push in no way affects your opponents future table image -- it only affects yours. Therefore you only have to track your table image. And the only real impact of your table image change is the change on opponents' calling ranges. Your opponents table image changing your calling range, again, doesn't matter.

[Nicholasp27]

it's incorporated into the model THAT hand

the next hand's ev is NOT in the model


do u agree with this statement:

if your action in hand X causes the opps' ranges to loosen in hand Y, then your ev will be less in hand Y than if you hadn't taken that action in hand X

if u agree, then do u agree that sometimes that difference in Y's ev can be more than .1? if so, then do u agree that it may be worth passing up a .1% ev move if it will affect your table image?



(and that's not even considering the fact that in real-world usage, you aren't accurate enough with your assessment of calling ranges to know that it really is + if it's +.1% ev)

[Nicholasp27]

table image of yourself for each opp

not talking about their numbers...saying that their number FOR YOU may vary from opp to opp