Originally posted by conall88
dont lie, ICM + reads
Originally posted by jbpatzer
I expect an exploitative strategy would do significantly better against both.To exploit and be exploitable, or not to be exploitable, that is the question!
Notifications
Clear all
Originally posted by jbpatzer
I expect an exploitative strategy would do significantly better against both.And I'm struggling to find anything much better or worse than straightforward ICM atm.
Do you mean better against wimp and maniac or a better TEQ estimation in general?
For the first one, i'd start deviating ranges from ICM in the easy-calculatable cases first, which are sb vs bb against maniac/wimp and being OTB with the wimp in the BB (because it's effectively blind vs blind against the ICM player with better odds).
- shove any 2 cards bvb against the wimp regardless of stacksizes
- shove and call the same range of hands bvb against the maniac and this includes all hands which have the minimum equity of
E = (TEQ(fold)-TEQ(lose))/(TEQ(win) - TEQ(lose)) against any 2 cards. The TEQ can be calculated by via ICM or M-W or whatever does somewhat ok.
- shove at least as many hands from the BU with the wimp in the BB as you would if you were blind vs blind against the ICM player.
I think this strategy should beat Maniac and Wimp better than ICM nash since you don't show up with too loose ranges against the ICM guy.
EDIT: One further adjustment should generate a lot of EV. Tighten up your BU-range a lot if the maniac is in the BB. According to ICM-nash, the ICM-SB will shove quite a lot. This means your two opponents will collide a big amount of the time. This is exactly what you want on the bubble. I could imagine shoving 0 hands in a spot like having 4bb, the ICM nash SB and the maniac BB having 13bb each.
If you mean the TEQ estimation: Yeah, if ICM has the equilibrium right for N=25, there's not much we can do. But you can at least say you measured "real TEQ" in your model by playing out ICM vs ICM vs ICM. If they all played the correct nash ranges, the outcome result has to be real TEQ.
Originally posted by muebarek
i'd start deviating ranges from ICM in the easy-calculatable cases first, which are sb vs bb against maniac/wimp and being OTB with the wimp in the BB (because it's effectively blind vs blind against the ICM player with better odds).
- shove any 2 cards bvb against the wimp regardless of stacksizes
- shove and call the same range of hands bvb against the maniac and this includes all hands which have the minimum equity of
E = (TEQ(fold)-TEQ(lose))/(TEQ(win) - TEQ(lose)) against any 2 cards. The TEQ can be calculated by via ICM or M-W or whatever does somewhat ok.
- shove at least as many hands from the BU with the wimp in the BB as you would if you were blind vs blind against the ICM player.I think this strategy should beat Maniac and Wimp better than ICM nash since you don't show up with too loose ranges against the ICM guy.
EDIT: One further adjustment should generate a lot of EV. Tighten up your BU-range a lot if the maniac is in the BB. According to ICM-nash, the ICM-SB will shove quite a lot. This means your two opponents will collide a big amount of the time. This is exactly what you want on the bubble. I could imagine shoving 0 hands in a spot like having 4bb, the ICM nash SB and the maniac BB having 13bb each.
All sensible suggestions, but I want an algorithm that adjusts the ranges of the other two players automatically. In other words, what's their equilibrium strategy, knowing that the other guy is a wimp or a maniac? Probably something like you suggest, but surely it should be possible to calculate it. I'm currently running something along these lines, and find that two ICM wimp exploiters do worse than standard ICM against a wimp! This could just be something wrong with my code, but funny things can happen in multiplayer games (maybe pzhon might have some pearls of wisdom on this), so I'll keep at it and report back. In any case, I think there's probably something interesting to be learnt by trying to find an algorithm to get the best exploitative strategy against wimps and maniacs.
Thanks Gabinr1 and Conall for the inspiration, whether you meant to be inspiring or not! 
Actually, the best strategy against a wimp is just to cooperate and let him blind out. Then the other two players just get 50% each. That's the problem with games with more than 2 people. Cooperation! Maybe if the two ICM players try to beat each other, they end up giving extra equity to the wimp. Very confusing, but interesting.
ok. that's right of course. but ingame you can only influence your own play. so you can't count on your opponents to adjust well, too. in game theory, one usually doesn't rely on cooperation. So my suggestions were based on the situation that you have the one maniac/wimp and one player who blindly pushes ICM nash and our goal is to make the best out of the situation. Actually, that's not a too unrealistic scenario (of course the maniac/wimp are highly exaggerated), since there are a lot of people hardcore multitabling using their autopilot "pushbot"-ranges rather blindly.
Edit: and the nice thing is: in this case (with only adjusting one player's ranges), the ranges should be reasonably easy to calculate (with exception of shoving the BU with the maniac in the BB where the future game aspect of having the other two stacks clash a lot has to be taken into consideration).
Edit2 ♦: If you're interested in the equilibrium of the two players, isn't this even easier to calculate than a three player equilibrium? Couldn't you just let find the max EV range (calculated by ICM) for the player last to act (of the two) against every possible range of the one who acts first and then pick the range which is max EV for the first player given the "perfect" response of the one last to act? This should work out since the fish's ranges are constant.
Originally posted by muebarek
ok. that's right of course. but ingame you can only influence your own play. so you can't count on your opponents to adjust well, too. in game theory, one usually doesn't rely on cooperation. So my suggestions were based on the situation that you have the one maniac/wimp and one player who blindly pushes ICM nash and our goal is to make the best out of the situation. Actually, that's not a too unrealistic scenario (of course the maniac/wimp are highly exaggerated), since there are a lot of people hardcore multitabling using their autopilot "pushbot"-ranges rather blindly.Edit: and the nice thing is: in this case (with only adjusting one player's ranges), the ranges should be reasonably easy to calculate (with exception of shoving the BU with the maniac in the BB where the future game aspect of having the other two stacks clash a lot has to be taken into consideration).
Edit2 ♦: If you're interested in the equilibrium of the two players, isn't this even easier to calculate than a three player equilibrium? Couldn't you just let find the max EV range (calculated by ICM) for the player last to act (of the two) against every possible range of the one who acts first and then pick the range which is max EV for the first player given the "perfect" response of the one last to act? This should work out since the fish's ranges are constant.
One wimp/maniac, one ICM, and one adjusting is a good idea. I'll try that.
And 'Edit 2' yes in principle, but there may be more than one equilibrium. But I could have a look.
This is a wimp (always folds) playing against two players who know ICM and try to exploit the fact that the wimp always folds.
The wimp does better than he did when he was up against two players pushing the Nash ICM ranges.
Why? Well, the two ICM players know that the wimp will be folding, so they adjust their ranges to exploit this by pushing wider, but they don't know that he is always going to fold. They still assign a high value to his stack, even though we know it is worthless, as he never uses it. They end up going for a short term gain, and leaking equity to the wimp.
Cool! 
Hm. Yeah. The wide shoving isn't the problem there, imo. The wide calling is. By overestimating the wimp's TEQ (which is zero) they undervalue their own equity - this means they underestimate their risk aversion which makes them call too light against the wider shoving ranges. Whenever the two are all-in, the wimp shows a profit.
To show that this is the case, one just has to let them push the exploitative ranges but only call ICM nash ranges. This should show a better result than ICM nash since you lose nothing when calling compared to ICM-nash but you'll show a better profit shoving!
EDIT: the easiest way to get a better result than nash ICM will be to just shove 100% bvb against the wimp and play according to nash ICM otherwise.
Originally posted by muebarek
Hm. Yeah. The wide shoving isn't the problem there, imo. The wide calling is. By overestimating the wimp's TEQ (which is zero) they undervalue their own equity - this means they underestimate their risk aversion which makes them call too light against the wider shoving ranges. Whenever the two are all-in, the wimp shows a profit.To show that this is the case, one just has to let them push the exploitative ranges but only call ICM nash ranges. This should show a better result than ICM nash since you lose nothing when calling compared to ICM-nash but you'll show a better profit shoving!
EDIT: the easiest way to get a better result than nash ICM will be to just shove 100% bvb against the wimp and play according to nash ICM otherwise.
Yes. Agreed. But the key point whatever way you look at it is that to play well the ICM players have to know the intentions of the wimp not just in the current hand but in future hands too. It's no good shoving 100% BvB when the wimp is BB if the wimp suddenly grows balls and starts playing the ICM Nash ranges. If the ICM players stick to the ICM Nash ranges, they're not open to exploitation by a sudden change of strategy. I have experienced this myself when someone has been disconnected (enforced wimpiness) and reconnected after I've pushed 32o BvB. Nasty!
And for completeness, doing the same thing for a maniac, it doesn't make a lot of difference from just playing straightforward ICM.
Originally posted by jbpatzer
Yes. Agreed. But the key point whatever way you look at it is that to play well the ICM players have to know the intentions of the wimp not just in the current hand but in future hands too.
exactly. that's what poker is all about. not only identifying and exploiting weaknesses but also recognizing if an opponent is capable of adapting and if yes, foreseeing how fast will he adapt and in what manner (overadaption, adapting in a wrong way, etc.).
easy game, huh? ♠_biggrin:
but seriously. this is the point where experience and psychology trump math - so the only way to be on the mathematically safe side (for a simple computer program for example, that can't really rely on this) will be playing in the game theoretical optimum which isn't the winrate optimum for sure ^^
OK. Agreed. So I think there's plenty of evidence that Nash ICM works well. What we can go back to is adjusting it for position. All of this has been for randomized positions. So I'm going to go back and read what you wrote about this earlier in the thread. Let's see if we can make any progress on that.
Originally posted by jbpatzer
What we can go back to is adjusting it for position. All of this has been for randomized positions.
And this is where it gets really interesting. Position adds so many new problems. Just take the wimp/maniac cases for example: I'd be surprised if the relative position on such an extreme player type wouldn't have significant influences on strategy.
I’d suggest we approach this in the way how we finished the random position problem. It will be of no use coming up with a great position dependent TEQ prediction model, if there’s actually nothing to improve in terms of nash ranges.
So my suggestion would be to find out, if there is a chance of beating ICM nash by deviating from the ICM nash ranges in a certain spot by +-1 hand. If there’s an improvement in one of the directions, add/subtract another hand and see if it gets even better – and so on…
So our main focus should be (since we don’t have infinite time ^^) to determine the spots where changing the ranges might make sense from a poker player’s point of view (for example shoving first in lighter with <=5bb to not lose your fold equity is an argument you hear from a lot of players)
Also, using a satellite structure will be sensible since we saw earlier that positional effects are amplified by increasing the second place payout.
Originally posted by Waiboy
I'd like to contribute constructively to this thread.
Go on then.
Originally posted by muebarek
So our main focus should be (since we don’t have infinite time ^^) to determine the spots where changing the ranges might make sense from a poker player’s point of view (for example shoving first in lighter with <=5bb to not lose your fold equity is an argument you hear from a lot of players)
Some human opponents call too widely when they get 2:1, and too tightly when they get 3:2. Against these opponents, you may want to push a lot at 5 BB, although not just UTG. The Nash equilibrium strategies may not have these human weaknesses, and they are much less universal now than they were a few years ago.
What I expect to see is that you gain ICM equity outside the blinds, and you lose equity in the blinds, but future gains and losses should be discounted rapidly when the blinds are large because there is a decent chance that players will be eliminated, changing the situation. This should mean that your true tournament equity is lower when you are about to post the big blind than the ICM predicts, and greater when you are about to be the button. The amount of the difference UTG should be a fraction of the cost of giving up the blinds, but that might be significant if the blinds are large relative to your stack.
A separate issue is whether your position relative to the big stacks or small stacks helps or hurts yous. You might not be able to find as many profitable pushes if you have to push through a microstack, or through the chip leader. However, if the chip leader is on your right, then he will push aggressively through you and you will not find many profitable calls. So, this is not obviously unbalanced.
Originally posted by pzhon
Some human opponents call too widely when they get 2:1, and too tightly when they get 3:2. Against these opponents, you may want to push a lot at 5 BB, although not just UTG. The Nash equilibrium strategies may not have these human weaknesses, and they are much less universal now than they were a few years ago.What I expect to see is that you gain ICM equity outside the blinds, and you lose equity in the blinds, but future gains and losses should be discounted rapidly when the blinds are large because there is a decent chance that players will be eliminated, changing the situation. This should mean that your true tournament equity is lower when you are about to post the big blind than the ICM predicts, and greater when you are about to be the button. The amount of the difference UTG should be a fraction of the cost of giving up the blinds, but that might be significant if the blinds are large relative to your stack.
I said consciously "first in", not UTG 
Hm. In our case (3way bubble) there is only one position out of the blinds. What do you think - will this even increase the urgency of shoving lighter than ICM nash on the BU since you only have this one hand until the blinds hit you again (given large blinds relative to BU's stack)? And if yes, at which stacksize would you suggest to start adding one hand to the shortstack's BU range (for a first try. ofc this will depend on the other stacksizes, too).
Originally posted by jbpatzer
Originally posted by Waiboy
I'd like to contribute constructively to this thread.Go on then.
Oh, that's it really.
Originally posted by Waiboy
Originally posted by jbpatzer
Originally posted by Waiboy
I'd like to contribute constructively to this thread.Go on then.
Oh, that's it really.
Your work here is done! 
I tried ICM v ICM v fish (fish = push 30%, call 15%, overcall 5%), and found that whether ICM was straight Nash ICM, or adjusting to exploit the fish, ICM won about 34.5% each and the fish 31%. This seems a bit surprising, but having seen the results for maniacs and wimps, perhaps it isn't. I think the problem is that the two ICM players have to respect each other, which limits the extent to which they can exploit the fish, and also that push 30%, call 15%, overcall 5% is perhaps not as fishy as all that. An occasional completely random spite call from the big blind would probably make it more fishily realistic!
As I posted in my blog, before moving on to non-random starting positions, I'm going to tidy up and comment my MATLAB code and make it available here. Then other people can try it. And by other people, I probably mean muebarek!