Quote:
Originally Posted by Brilvis
Hi plexiq,
I'm loving your software and I also loved reading your thesis - I think I've read it about six times at this point! I have a few short questions which I was hoping you could answer for me.
Glad to hear you found it interesting
Quote:
(1) ICMIZER just introduced FGS in their multitable ICM simulations. Any chance such a feature will also be included in HRC at some point?
MTT FGS is relatively trivial to add, assuming we leave stacks from the other tables static over the course of the calculation. But leaving other tables static seems problematic for spots that are most interesting, e.g. bubble situations, that's why we haven't enabled it.
If you run a bubble spot with e.g. FGS-3 and static MTT stacks then you are essentially calculating ranges under the assumption that no player will bust on other tables within the next 3 hands. This may easily be worse than just using regular ICM for that situation.
Does ICMIZER use static stacks for MTT FGS or do they simulate play on other tables as well?
Quote:
(2) What does the feature 'Full optimization beyond the first level' mean in the FGS preferences? It notes that "if checked, the above settings [using ICM estimates for split pots] only apply to the first level", which to me sounds like a negative thing, and yet it is a recommended option. What does this option do in terms of accuracy and speed?
If you check this option, all of the other optimizations will be active from level 2 forward. This means slightly lower accuracy / higher speed if you leave the depth constant, but it may allow you to run higher depth for increased accuracy overall.
The split pot options should essentially always be checked, you'd need to carefully craft scenarios where split pots make any significant difference to the EQPosts. Splits frequently matter for individual hands, but that's not relevant for the FGS equity calculations. (We may actually just get rid of the split options and always enable the optimizations.)
The general 3-way option is more of a judgement call, if 3-way all-ins happen fairly frequently in your calculated hand then you may want to leave this optimization disabled.
Quote:
(3) Data question: does your software collect/store simulations done through your software and/or website?
The only user-calculations we store on our end are hands that are explicitly uploaded via "Hand: Share Hand". Results from website calculations are temporarily cached in-memory to improve performance, but not on permanent storage.
Quote:
(4) And, lastly, more out of curiosity, did you ever run extensions of your Master's thesis (like six player games, or larger stack averages), and, if you did, at what point did reasonable FGS depths start deviating substantially from equilibrium?
Thanks a lot!
We did "full tournament" bounty solves with the thesis framework back when the bounty models were initially added to HRC, to decide on an appropriate bounty model. Didn't run anything with 6+ players or larger stacks though, no.
But I've actually started to dust off the old thesis code recently to get it ready for new runs, especially with PKO.