Holdem Resources vs. ICM Trainer

    • prevolsek
      prevolsek
      Bronze
      Joined: 18.02.2009 Posts: 289
      Hy guys I am now starting to learn how to play SNGs. I played cash games before. I decided to learn the NASH ranges by heart from almost every position for almost all stack sizes.

      Now I know how NASH is calculated but I have a problem. When I compare the results from the online HoldemResources calculator and the results from the ICM Trainer I get different results...

      For example HR suggests that I should call from the BB with a range of 68,6% when HU against the SB (chip leader) with 6 BBs. ICM trainer suggest that I should call with 62% in the same spot. Heros edge in ICM Trainer is set to 0%.

      Does anyone understand why the calculations aren't the same? Also what should I trust more, the calculation with HR or the one with ICM Trainer?

      Thanks in advance
  • 6 replies
    • prevolsek
      prevolsek
      Bronze
      Joined: 18.02.2009 Posts: 289
      Ok I just found the answer on the HR website:

      If you own a copy of Chen/Ankenmans "Mathematics of Poker" (highly recommended, btw.), or SnGPT, you may have noticed that their heads-up NE push/fold solutions differ slightly from the ranges calculated by the ICM Calculator. To allow efficient calculation, the range construction is constrainted to "linear ranges" in this implementation.

      You can find charts with the optimal HeadsUp push/fold strategies here. These charts are calculated without handranking constraints.

      I don't understand it completely though, so if someone could elaborate i would be grateful.

      I did check if this was true and yeah the results from similar situations with both calculators is much more accurate but there are still deviations anyway.

      So still which do you trust more ICM Trainer or HR calculator?
    • wiarygodny
      wiarygodny
      Bronze
      Joined: 16.07.2009 Posts: 1,395
      here's a thread ive started long time ago when i had similar questions to those you have now. Maybe you can have a look and find some answers youre looking for. If sth is still not clear than post it here and ill try to help as long as i know the answers (although im definitely not an expert on those tools, just use them a lot myslef so have some experience let's say :)
    • prevolsek
      prevolsek
      Bronze
      Joined: 18.02.2009 Posts: 289
      Thanks a lot wiarygodny, that was very helpful. I'm still not 100% satisfied with what I found out to be honest, but still as expected both tools are helpful and I will continue to use both.
      But the fact that I know that the calculations differ bugs me just like it bugs you. :)
    • wiarygodny
      wiarygodny
      Bronze
      Joined: 16.07.2009 Posts: 1,395
      glad i could have helped, even if its still not 100% clear for you. btw. whats that exactly? maybe i can help further, at least try?

      The guy that youve seen posting in that thread, plexiq is actually the guy behind the holdemresources.net from what i know so maybe you can try asking him directly.
    • plexiq
      plexiq
      Diamond
      Joined: 28.12.2006 Posts: 1,045
      Sorry, only saw this thread now. If you have questions regarding holdemresources, it is best to use the support threads in the software forum. I have mail notifications setup for these.

      Originally posted by prevolsek
      For example HR suggests that I should call from the BB with a range of 68,6% when HU against the SB (chip leader) with 6 BBs. ICM trainer suggest that I should call with 62% in the same spot. Heros edge in ICM Trainer is set to 0%.

      Does anyone understand why the calculations aren't the same? Also what should I trust more, the calculation with HR or the one with ICM Trainer?
      The "core" calculations of the Web Nash Calculator and the ICM Trainer are still very similar. The ICM Trainer uses an old .Net port of the holdemresources.net engine (2007-2008).

      I don't know exactly how that core library is used in the trainer UI, but here is my best guess about the differences:

      Both the trainer and the online calculator use linear ranges while performing the Nash Equilibrium calculations. (ie, the ranges are restricted by a hand-ranking)

      In both cases, these linear ranges are never directly displayed to the user, but first optimized to improve hand selection.

      Now there are two basic approaches to do that:
      #1) Select hands with >0 EV
      #2) Select the top x% of hands by EV, such that the % of hands is equal to the originally calculated linear range

      I believe #1 is what the trainer does, and that is what holdemresources used to do as well. I switched the online calculator to use #2 a few years back, because the displayed ranges are usually much closer to a NE than #1. But more importantly, #1 can result in some *really* bad ranges in extreme cases, while #2 is comparatively robust.

      e.g.:
      Lets assume the NE calculation results in a linear calling range of 30%, but 10% of hands actually happen to be slightly -EV (not that uncommon when using linear rankings).

      If you display the range using #1 you may get a ~20% range, and now the calculated pushing ranges will not make sense in relation to this tighter calling range.

      With #2 you still get a 30% range, you just optimize the selection of those 30%. (This will still mess with the pushing range EVs a bit, but not nearly as bad as #1.)
    • prevolsek
      prevolsek
      Bronze
      Joined: 18.02.2009 Posts: 289
      Thank you plexiq and wiarygodny!

      Now I think I found out what I wanted. I think... :D

      So Plexiq you're trying to say that ICM Trainer and HR Calculator both caculate the same result. Then ICM Trainer just gives you a result of say 30% and shows you the top 30% hands equity wise and these 30% of hands are always the same no matter what the situation is?

      On the other hand HR adapts these 30% of hands even further, meaning that a 30% push from UTG will have different hands included then a 30% push from the SB because the calling ranges will be different and induvidual hands will have different HEV from these positions against different REV?

      If that's the case then that's great since now I know how I need to continue to study the game :s_cool: