Author: Dan Homan
Date: 05:51:35 06/04/99
Go up one level in this thread
On June 03, 1999 at 15:58:10, Inmann Werner wrote: >On June 03, 1999 at 14:20:00, Dan Homan wrote: > >>On June 03, 1999 at 11:37:00, Inmann Werner wrote: >> >>>Hello >>> >>>the "hit ratio" of hash tables confuses me a bit. What is a hit? >>> >>>I tried the LCTCMB01 position and got >>> >>>1) 1.520.234 nodes >>>2) 263.041 entries in Hash table used >>>3) 1.458.957 probes in Hash table >>>4) 137.961 entries found >>>5) 40.777 entries have a "allright" depth to use >>>6) 37.517 match alpha, beta (upper and lower fails) and are used! >>> >>>so i get a effective hit ratio of 37.517/1.458.957 =2,6% ???? >>> >>>where do the 10% going around come from??? >>> >>>can anybody explain, how he counts and what numbers he get? >> >>I think that most people are counting like you do. It depends on >>whether you store/probe frontier or q-search nodes. I never store >>or probe for q-search nodes, but I have tried it both ways with >>frontier (depth = 0) nodes. When I probe but do not store frontier >>nodes, I find that I get about 1% to 5% efficiency in most positions. >>When I probe and store frontier nodes, I get 10% or better efficiency, >>but this seems slower overall than only probing frontier nodes. >> >> - Dan >> > >I do not differ between q-search an normal search. But if i only count "normal >search" i get about 10% real hits, but it depends very much on search depth. >Only "normal search" >7 plys 4% >8 plys 8% >9 plys 10% > >Is this normal rates? I don't know. Every program/implementation is different. I think that your hit percentage should increase somewhat as you search more (unless you saturate your hash table), but it seems a little funny to me that you should double your hit percentage in going from 7 to 8 plys. Do you have this effect in a large number of positions? - Dan > >Werner
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.