Quote:
Originally Posted by LPump
Do you know in what direction the optimizations are - are they for the speed/size of the trees or for the accuracy? Do you think its a good idea to rerun older trees with the new algorithm?
Haven't seen any details described, except optimizations are mentioned in the changelog. Speed improvements are huge for the trees I've rerun (2-3x faster for a preflop tree I tested).
When comparing a new vs old flop sim I saw that buckets/node for flop was lower in the new version (but buckets/node for turn and river were identical). Not sure exactly how these settings change accuracy though. RAM use is identical from what I have seen.
I'm keeping a backup folder of v 10106, but I'll work with updated version from now on. If discrepancies are due to an improved algorithm we should of course rerun old sims, but I'll hold that off for now. Would be nice if developer could comment on that, but that seems unlikely (seeing him still working on the software is good enough for me).
I did test Monker vs Pio for a HU case a long time ago, and was happy with the results (virtually identical), so I'll trust the algorithm until something is demonstrated to be bogus.
Last edited by ZenFish; 10-03-2019 at 04:17 PM.