Author: Omid David
Date: 15:40:24 07/11/02
Go up one level in this thread
On July 11, 2002 at 18:35:04, Uri Blass wrote: >On July 11, 2002 at 18:20:51, Omid David wrote: > >>On July 11, 2002 at 18:04:37, Uri Blass wrote: >> >>>On July 11, 2002 at 17:46:36, Omid David wrote: >>> >>>>On July 11, 2002 at 17:41:08, Uri Blass wrote: >>>> >>>>>On July 11, 2002 at 16:38:50, Omid David wrote: >>>>> >>>>>>As part of an extensive research (will be published soon), we tested null-move >>>>>>pruning with fixed depth reductions of R=2 and R=3 on about 800 positions of >>>>>>"mate in 4" (searched to depth of 8 plies) and "mate in 5" (searched to depth of >>>>>>10 plies). The results naturally show that R=2 has greater tactical performance >>>>>>(greater number of checkmate detection). >>>>> >>>>>This is not the right test. >>>>>It is clear that if you search to fix depth R=2 is going to be better. >>>>> >>>>>The question is what happens when you search for the same time. >>>>> >>>>> >>>>> However, we also conducted about >>>>>>hundred self-play matches under 60min/game time control between R=2 and R=3. The >>>>>>outcome is a rather balanced result (R=2 only a little better). Considering that >>>>>>the tests where conducted on a rather slow engine (100k nps), on faster engines >>>>>>R=3 is expected to perform better. >>>>> >>>>>I think that a better test should include different programs and not the same >>>>>program against itself. >>>>> >>>>>Another point is that R=2 and R=3 are not the only possibilities. >>>>>> >>>>>>So, apparently R=2 is not _by_far_ better than R=3 as some assume. >>>>> >>>>>I suspect that it is dependent on the program(results may be different for >>>>>programs with different qsearch and different evaluation). >>>>> >>>>>Uri >>>> >>>>I don't expect R=2 to gain more from greater speed than R=3. As a matter of fact >>>>as Dr.Hyatt recently mentioned with faster hardware in the future, R=3 might >>>>reach depths in which the total saving would be more significant than tactical >>>>deficiency (deeper search would compensate for it). In such cases one might even >>>>think of R=4 at some parts of the search tree (or as Dr.Hyatt just mentioned an >>>>adaptive R=3~4 value). >>> >>>I agree that R=3 and R=4 should be also considered. >>> >>>From my experience in GCP test suites at 5 minutes per move recursive R=3 got >>>the best results so far in GCP test suites(I do not know about games because I >>>did not test it for games at slow time control but I suspect that it is possible >>>that for movei R=3 may be even better than R=2 at the same time control). >> >>I think in blitz time controls R=2 will be by far stronger than R=3, since the >>search won't get deep enough and consequently: >>1. R=3 won't save too much search effort in comparison to R=2 >>2. R=3 won't have time to search deeper to compensate its tactical deficiencies. >> >>But in deeper time controls especially in faster engines like Crafty, it'll >>perform better I believe. > >1)Crafty is not using R=3 but R=2/3 so comparison between R=3 and R=2 is not >relevant for crafty. But you can set a fixed R-value by setting the MAX and MIN to the same value. It might be interesting to conduct Crafty self-play with R=2 vs. R=3. > >2)Crafty is not doing checks in the first plies of the qsearch and I guess that >the result is bigger advantage for R=2 in blitz. > >Uri While working on test suites I delibertely turn off checks in qsearch, to get better distinction between different R-values.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.