Author: James Swafford
Date: 08:22:30 09/18/02
Go up one level in this thread
On September 18, 2002 at 11:14:14, Robert Hyatt wrote: >On September 17, 2002 at 23:32:03, James Swafford wrote: > >>On September 17, 2002 at 23:03:13, Robert Hyatt wrote: >> >> >>>> fail highs: 10866 (3%); fl: 9929 (3%); ex: 671 (0%) >>> hashing-> 29%(raw) 24%(depth) 99%(sat) 99%(pawn) >>> hashing-> 0%(exact) 16%(lower) 1%(upper) >>>> >>>> fh: 1531962 (49%); fl: 432594 (13%); ex: 2836 (0%) >>> hashing-> 64%(raw) 59%(depth) 99%(sat) 99%(pawn) >>> hashing-> 0%(exact) 55%(lower) 2%(upper) >>>> >>>> fh: 3443 (1%); fl: 2926 (1%); ex: 151 (0%) >>> hashing-> 24%(raw) 18%(depth) 98%(sat) 98%(pawn) >>> hashing-> 0%(exact) 16%(lower) 0%(upper) >>> >> >>Thanks Bob. I've cut and pasted your data to follow mine for >>each position. I don't count the number of probes I do.. maybe >>I should. Anyway, on positions 1 and 3 I noticed your fail high% >>is significantly higher than mine. On position 2, my fail low% >>is significantly higher than yours. Position 2 is more "tactical". >> >>How do you interpret that? Is it likely that I have a bug, or >>is this a consequence of poor (or different) move ordering? >>My move ordering is horrible right now... but I'm not sure how that >>affects my hash table usage. > > > >That is a tough thing to compare. IE do you probe in q-search? I don't. >That might make the comparison a bit flakey. fail high vs fail low could >certainly vary, although they probably should add up to a similar total. No, I don't. > >IE I can fail high at the current ply on a probe, you could fail low on >the next ply and cause the same sort of tree size... > >In tactical positions, extensions play a role, of course. Which would lead >me to not pay a lot of attention to comparing that... I don't have PVS implemented yet... just a modest aspiration window at the root. I think that accounts for some difference (after seeing Andrew's numbers). > >also, 10 second searches might be deeper or shallower for me, which would >also affect hashing efficiency. You're much deeper, I'm sure. > > > > >> >>BTW - I'm using a combination of depth preferred and always >>replace. The "tables" are the same size (really one big table >>so I can do consecutive reads). > >For PC performance that is probably better, although the 1x / 2x tables >I use let me use 3/4 of memory for hash, rather than 1/2... > > >> >> >>-- >>James
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.