ICM and $EV lines

    • IhasLuck
      IhasLuck
      Bronze
      Joined: 07.05.2010 Posts: 326
      I've had some thought about ICM and it's imperfection, this is what I came up with..
      ICM is used as the main model for $EV calculations in tracking programs as far as I know and the model itself has some flaws in it. So I think that makes the $EV lines the tracking programs inaccurate for some players.
      The players who find spots that are Actually more +$ev (based on results) than what calculations based on icm dictate over a huge sample would see that their $won graph is lower than their $ev graph.
      And if other people avoid spots that are Actually -$ev when ICM would suggest that they're profitable again over huge sample would see that their $ev graph is higher than their $won graph.
      What I'm saying that depending on the player and how he plays from his experience,intuition,etc.. due to the imperfection of ICM the $ev graph in the tracker and his actual expected results may vary.

      What are your thoughts on this?
  • 11 replies
    • pzhon
      pzhon
      Bronze
      Joined: 17.06.2010 Posts: 1,151
      If the equities estimated by the ICM are wrong, this just means the variance reduction does not work as well. Maybe it takes 600 instead of 550 reduced results to be as accurate as 1000 unadjusted results. If inaccurate equities are used in the correct luck adjustment formula, this does not introduce a bias.

      This may be counterintuitive, but that's how the math works out, both in poker and elsewhere. A similar counterintuitive truth is that whether your playing style leads you to get your money in while ahead or behind does not affect whether the luck estimate will tend to be positive on average.
    • BOBOsvk
      BOBOsvk
      Bronze
      Joined: 14.02.2010 Posts: 2,505
      ev lines suck, i just lost 250$ on ev in a 100$ sng :D
    • VorpalF2F
      VorpalF2F
      Super Moderator
      Super Moderator
      Joined: 02.09.2010 Posts: 8,904
      The EV lines are misleading.
      They can only actually calculate EV when the other players' hands are known.

      You can make horrible shoves, and get folds, or make perfect shoves and get beaten, by a caller who made a perfect call, or a bad call.

      I think the only purpose of the EV line is to make you feel better when you lose.
      As in, "Well I lost, but it was just luck" etc.

      It is even more misleading in cash games.
    • IhasLuck
      IhasLuck
      Bronze
      Joined: 07.05.2010 Posts: 326
      Originally posted by pzhon
      If the equities estimated by the ICM are wrong, this just means the variance reduction does not work as well. Maybe it takes 600 instead of 550 reduced results to be as accurate as 1000 unadjusted results. If inaccurate equities are used in the correct luck adjustment formula, this does not introduce a bias.
      What do you mean by correct luck adjustmen formula?
      Vorpal that's obvious, but the point is that real results are more misleading than EV.
    • pzhon
      pzhon
      Bronze
      Joined: 17.06.2010 Posts: 1,151
      I mean that the correct method is used: You add a quantity which averages to 0. The weighted average of the luck adjustment when you win and the luck adjustment when you lose has to be 0, when weighted by the probabilities of each. This doesn't change the mean.

      If your equity estimates are really accurate, you reduce the luck-while-all-in component of the variance almost to 0, but of course there are other sources of variance. If your equity estimates are not so accurate, your reduction for the luck-while-all-in component of the variance is less impressive, but this doesn't introduce a bias. The math for this is simple, and it has also been tested.

      This doesn't mean that the luck adjustment is actually unbiased. There were some bugs in some implementations, including when programs were asked to analyze multitable SNGs. There is still a small card removal effect which means that the unseen cards are not uniformly random because the actions of the other players who folded may indicate that more good cards are left in the deck. But there is no bias just because the ICM is imperfect, and you might know how to play better than the ICM says in a few situations.
    • evertonroar
      evertonroar
      Bronze
      Joined: 26.06.2009 Posts: 737
      meh. if i make a +ev shove and villian folds, there is no way known that HEM or PT3 can calcualte the ev.

      If villan calls sometimes I will be ahead and sometimes I will be behind. Then sometimes I will win and sometimes I will lose. if the range I put him on is correct the shove is still +ev. But hem will show it as +ev won or lost depending on the result. This is the only time Hem can calculate ev.

      if I shove KK with 7BB against a 8 BB stack on the bubble and he shows AA and I lose, it will be slightly ev+ (cos i got a little unlucky that I didnt win). If I win it will show a huge negative impact on my ev as i sucked out. either way i made the right decision.

      As far as im concerned, if you are pushing correctly, your actual winnings will be above the ev in the long run as you are winning: what the ev lines shows + all the winnings from the postive ev shoves they folded. (minus a small amount thats taken into account becuase your stacks larger when calcuating ev when they call).
    • pzhon
      pzhon
      Bronze
      Joined: 17.06.2010 Posts: 1,151
      That's a common misunderstanding of luck adjustment. Do the calculations in some examples and you will see that adjusting for the all-in luck makes the adjusted results more accurate without introducing biases.

      Suppose you get all-in with KK vs. AA. Does your playing style affect your chance to win this hand now? No. Does whether your shove was correct affect your chance to win this hand now? No. So, programs like HEM don't need to understand your playing style, or to figure out whether your shove was correct, to estimate the part of your luck which occurs after you get all-in. Subtracting this part of the luck simply makes the adjusted results more accurate (lower variance) than the unadjusted results.
    • IhasLuck
      IhasLuck
      Bronze
      Joined: 07.05.2010 Posts: 326
      Originally posted by pzhon
      I mean that the correct method is used: You add a quantity which averages to 0. The weighted average of the luck adjustment when you win and the luck adjustment when you lose has to be 0, when weighted by the probabilities of each. This doesn't change the mean.

      If your equity estimates are really accurate, you reduce the luck-while-all-in component of the variance almost to 0, but of course there are other sources of variance. If your equity estimates are not so accurate, your reduction for the luck-while-all-in component of the variance is less impressive, but this doesn't introduce a bias. The math for this is simple, and it has also been tested.

      This doesn't mean that the luck adjustment is actually unbiased. There were some bugs in some implementations, including when programs were asked to analyze multitable SNGs. There is still a small card removal effect which means that the unseen cards are not uniformly random because the actions of the other players who folded may indicate that more good cards are left in the deck. But there is no bias just because the ICM is imperfect, and you might know how to play better than the ICM says in a few situations.
      That's awesome, I still don't get all the whys but I'll take your word for it. Very good read
    • reslodrek
      reslodrek
      Basic
      Joined: 21.08.2011 Posts: 4
      Let's assume that a player shoves MORE EV+ then the random players in the games. The player should make more money then the all in ev graph because the program can only estimate the all in EV for hands getting to showdown. The player in this case will make more money then EV0 in hands without showdown and therefore make more money then the AIEV graph will show? Am I wrong?:/
    • pzhon
      pzhon
      Bronze
      Joined: 17.06.2010 Posts: 1,151
      Originally posted by reslodrek
      Let's assume that a player shoves MORE EV+ then the random players in the games. The player should make more money then the all in ev graph because the program can only estimate the all in EV for hands getting to showdown. The player in this case will make more money then EV0 in hands without showdown and therefore make more money then the AIEV graph will show? Am I wrong?:/
      Sorry, you are wrong.

      The adjustments the program makes average to 0. When you add something with an average value of 0 to your results, the average doesn't change. So, both your results and your adjusted results have the same expected value. You do not expect to have higher results than adjusted results based on your playing style or skill advantage.

      Whether your push was correct or not, whether you are outplaying your opponents or not, does not affect your chance to win the hand once you get all-in. That's all that is necessary for the all-in luck adjustment to have average value 0. It is not necessary for the adjustments to estimate all luck, or even to be good estimates, to have an average value of 0.
    • reslodrek
      reslodrek
      Basic
      Joined: 21.08.2011 Posts: 4
      Ok thank's for the respond! I haven't been looking very much on EV graphs at all since i've always been a litle skeptical about them. Anyways, quick question; What program has the best EV features for sngs in your opinion?