Author: Tim Foden
Date: 04:00:30 12/19/00
Go up one level in this thread
On December 18, 2000 at 17:25:38, Scott Gasch wrote: >Realized after I sent last message that these might help you out a little too: > > Hash probes: 357103 Hash stores: 159816 > hit exact: 2084 store exact: 1166 > hit upper: 13164 store upper: 52337 > hit lower: 164747 store lower: 106313 > hit unusable: 101410 store nulllw: 0 > deep/fast: 257292 / 68267 (72.05% / 19.12%) > Efficiency: 91.17% > >Probes is calls to HashLookup. Hit exact is exact hits with enough draft to >use. Hit upper/lower is enough draft and usable with current A/B to cause a >cutoff. Unusable is not enough draft (but still ordered first). > >Stores are exactly what you think. Store nulllw is lower bounds stored after a >null move cutoff. The deep/fast numbers are for the two halves of the >transposition table... one implements deeper-only replace and the other always >replaces. These tables are the same size currently although I would be >interested in anyone with experience shrinking the always replace table... The >efficiency numbers include unusable hits (basically if I match a sig in the hash >the numerator of the efficiency ratio increases. The denominator is the total >probes). These numbers are from the output of that same 5 sec search I am >following up to. > >Good luck! >Scott Hi Scott, I have added some output to show hash table stats in my current version. This may have shown something... From a 5 sec search: trans: probes=383355 hits=381465 (99.51) draft=376184 (98.13) trans: exact=501 (0.13) upper=19463 (5.08) lower=1424 (0.37) From a 30 sec search: trans: probes=2155300 hits=2151794 (99.84) draft=2144458 (99.50) trans: exact=662 (0.03) upper=69608 (3.23) lower=4388 (0.20) As you can see, although GLC is getting lots of hits (>98%), it is getting nowhere near the number of cutoffs (<6%). In particular the number of lower bound cutoffs is very small compared to yours. This may well be getting me on to the right track... thanks for your help. Cheers, Tim.
This page took 0.01 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.