Push ranges in different sources

    • Avatars91
      Avatars91
      Bronze
      Joined: 18.12.2009 Posts: 2,689
      After having tried ICM trainer, SNG wizard and holdemresources.net Nash calculator I have discovered that the ranges they offer in identical situations differ and sometimes greatly. What I mean by this is that the percentage they suggest stays more or less the same or is at least close, whereas the hands that fall into these ranges are far from being similar.

      If I am not mistaken, then SNG wizard and HR.net Nash calculator's ranges were close if not identical, but the difference between a 35% range in ICM trainer and 35% range in Nash calculator is very notable.

      Thus I am very mistaken about which of these sources to trust. It also makes me doubt my current understanding of what to push and when. Why is there such a difference and which is the credible source?
  • 11 replies
    • pzhon
      pzhon
      Bronze
      Joined: 17.06.2010 Posts: 1,151
      I'm surprised to hear that there is a big difference between ICM Trainer and the Nash calculator. I thought they were using very similar algorithms. Please post examples.

      SNG Wizard uses pushing ranges which look too much like calling ranges. You can call slightly wider against a player who is pushing some weaker hands and folding some stronger ones. However, I don't think this makes a large difference in most situations.

      In most cases, you should not be obsessed with whether you are making all of the +EV pushes and folding the -EV hands. When a hand is marginal, it's not important whether you push or not. What is important is that you push the hands which are right by a lot, and fold the hands which would be very unprofitable. If programs disagree on marginal hands, that's not important. It's only a big deal if they disagree on hands where one says pushing or folding is right by a lot.

      A possible problem is that you might not have set the minimum edge to 0%. You should have a clear reason specific to the hand and your reads in order to use an edge other than 0. The default edge will give you strange results which may damage your game if you follow them.
    • Avatars91
      Avatars91
      Bronze
      Joined: 18.12.2009 Posts: 2,689
      There is a difference in the suggested ranges both in terms of percentage (depending on the situation it's either very small or somewhere around 5%, sometimes 10% though) and the hands that fall into certain %.

      For example, put an identical situation both in the Nash calculator and ICM trainer:

      Blinds = 150/300

      Position Stacksize

      MP3 3000
      CO 3000
      BU 4500
      SB 1500
      BB 3000


      Nash calculator: ICM Trainer:

      MP3: 21.00% MP3: 20.97%
      CO: 26.70% CO: 26.40%
      BU: 50.80% BU: 54.15%
      SB: 65.60% SB: 66.82%

      The difference in ranges percentage-wise is in this situation almost not noticeable, except for the ~3% difference regarding BU ranges. Take note that in other situations there might be bigger difference.

      What is noticeable, however, is the following:

      Nash calculator: 21.0%, 66+ A2s+ A9o+ A5o K9s+ K7s KTo+ Q9s+ QTo+ J9s+ T9s
      ICM Trainer: 20.97%, 22+ A3s+ A9o+ K8s+ KJo+ Q8s+ QJo J8s+ T8s+ 98s

      I lack the knowledge to assess, whether it is a big difference or not, but seeing such a gap between 22+ and 66+ makes me concerned. And in a sense this makes me doubt that any of these programs are correct. Well, obviously they seem not to be flawless, at least one of them.
    • pzhon
      pzhon
      Bronze
      Joined: 17.06.2010 Posts: 1,151
      I think they are disagreeing on marginal hands. They may use different numbers of iterations or different starting ranges. It's like one approximation algorithm is getting 0.98 while another is 1.007 when the actual answer is 1. I think you see similar differences between the Nash calculator and the more advanced Nash calculator beta.

      Some push/fold equilibriums are not very stable, so there might be quite a few hands with close to 0 equity. Changing the calling range slightly may change several hands from slightly negative to slightly positive or vice versa. The important thing is to get the nonmarginal hands right.
    • Avatars91
      Avatars91
      Bronze
      Joined: 18.12.2009 Posts: 2,689
      Could you please illustrate how big the difference between marginal and nonmarginal hands is?

      Also, I don't exactly understand why the results are different depending on the source because the mathematics and calculation regarding ICM should be identical in my opinion.
    • pzhon
      pzhon
      Bronze
      Joined: 17.06.2010 Posts: 1,151
      Set up a situation in SNG Wizard where the weakest offsuit king to push is K5o. Compare pushing K4o with K6o. Usually there won't be a huge difference, particularly if 44 and 55 are not in the calling range. For example, I still had the equal-stacks 12.5 bb bubble situation set up. K5o was the weakest pushing hand recommended. Pushing K6o was worth 0.01%. Pushing K4o cost 0.02%. If you fold K6o, or push K4o, when the correct boundary is K5o, these aren't large mistakes, and there are far more important things to study. If you fold A5s when pushing should gain 0.30% of the prize pool, or about 2.5% of a buy-in, this could be a significant mistake worth fixing.

      The Nash calculator at HoldemResources.net takes web queries. ICM Trainer is run on a PC. These might use the exact same parameters and the exact same algorithm, but they probably don't. If I ask two people to approximate the square root of 3 as a decimal, I don't expect to get the exact same answer because they are going to give me approximations. One might say 1.72, another might say 1.732. Both are useful and reasonably accurate. Since the various Nash calculators are approximating the Nash equilibrium, rather than computing the exact answer, they are just trying to get close, and they will differ on some marginal hands.
    • Avatars91
      Avatars91
      Bronze
      Joined: 18.12.2009 Posts: 2,689
      Thank you. As always, you explain everything excellently.

      As I don't want to make a separate thread on this, I would like to ask a question:

      How to assess an action and its profitability money-wise when pushing, i.e. if an action in a 10$ SNG has an equity of 0.30% (I might not have chosen the right words, but I hope you will understand what I'm saying), how much does it account for in dollars?
    • pzhon
      pzhon
      Bronze
      Joined: 17.06.2010 Posts: 1,151
      Originally posted by Avatars91
      How to assess an action and its profitability money-wise when pushing, i.e. if an action in a 10$ SNG has an equity of 0.30% (I might not have chosen the right words, but I hope you will understand what I'm saying), how much does it account for in dollars?
      The percentages reported by SNG Wizard are percentages of the prize pool. In a 9-player $10+$1 SNG, the prize pool is $90, so 0.30% would be 0.0030 x $90 = $0.27. One quick conversion is that multiplying by 8 approximates the percentage of a buy-in. $0.27 is about 2.4% of an $11 buy-in. If you fix a mistake like this in every tournament, your ROI should increase by about 2.4%. Of course, if you would only make a similar mistake once every 10 tournaments, correcting this tendency will improve your ROI by only 1/10 as much.
    • Avatars91
      Avatars91
      Bronze
      Joined: 18.12.2009 Posts: 2,689
      Thank you a lot! This is invaluable information!
    • Groging
      Groging
      Bronze
      Joined: 15.09.2010 Posts: 19,010
      pzhon , hello!
      Can you say, what we are must doing to calculate Nash ranges in all situations as on HoldemResorses.
      maybe there is some place with theory of Nash equilibrium decision or source of computer program?

      Thank you.
      Sorry for my bad English, i`m from russian community :f_cool:
    • pzhon
      pzhon
      Bronze
      Joined: 17.06.2010 Posts: 1,151
      Hi, Groging. I doubt anyone has made the source code for a Nash equilibrium calculator available. You may want to contact Helmuth Melcher, the creator of holdemresources.net. I believe he had an article in which he outlined the algorithm for approximating the Nash equilibrium, but I no longer see that on the holdemresources site. He is using a different method in the beta version of the Nash calculator.
    • Groging
      Groging
      Bronze
      Joined: 15.09.2010 Posts: 19,010
      Originally posted by pzhon
      Hi, Groging. I doubt anyone has made the source code for a Nash equilibrium calculator available. You may want to contact Helmuth Melcher, the creator of holdemresources.net. I believe he had an article in which he outlined the algorithm for approximating the Nash equilibrium, but I no longer see that on the holdemresources site. He is using a different method in the beta version of the Nash calculator.
      thank you!

      I research this problem(Helmut Melcher said in article that he used iterative method & twoplustwo poster said about algorithm that approximating Nash Equilibrium in heads-up )

      Generalize i`m now know how implement it to more-than-2 players-SNGs(i dont know how say it in English :) ) and write program now.

      Pzhon, thanks :f_biggrin:


      P.S.

      This post has been edited 1287 time(s), it was last edited by pzhon: Today, 01:04.

      :D :D :D

      funny signature :)