Author: Andrew Williams
Date: 15:32:56 05/03/01
Go up one level in this thread
On May 03, 2001 at 17:53:54, Dann Corbit wrote: >On May 03, 2001 at 17:50:18, Ulrich Tuerke wrote: > >>On May 03, 2001 at 17:30:25, Dann Corbit wrote: >> >>>On May 03, 2001 at 17:06:19, Ulrich Tuerke wrote: >>> >>>>On May 03, 2001 at 16:54:17, Dann Corbit wrote: >>>> >>>>>With this subset of 28 WAC records, could anyone who is willing run their chess >>>>>engine against this set: >>>>>ftp://cap.connx.com/pub/criminals.epd >>>>> >>>>>And tell me what percentage of correct first move is recieved by their program? >>>> >>>>What are you talking about ? >>>>Do you mean the cases where the best move hadn't changed over all iterations ? >>>>These cases will be very rare, I guess. >>>> >>>>Not sure, if I got you right, Dann ? >>> >>>What I mean is, you made 50,000 first guesses as to what was the move to analyze >>>first, and 45,000 were correct. >>> >>>That gives you 90%. >>> >>>Like that. >> >>Let me put it this way. >>In all cases, where a new best move had been found, we check if it had already >>been on 2nd place. If yes, we count it as a hit. >> >>The corresponding rate would be a measure of our root move ordering quality. > >When I run comet on and EPD suite, where will I find that measure? > >Crafty posts fail high percentage. I suspect that this measure is also nearly >identical, but I might be way off-base. Am I? Are you talking just about ply 1 move ordering, or are you interested in the overall figure? My program calculates the percentage of time that a move which gives a cutoff at a node (any node) was the first move that was considered at that node. Is this what you mean? If so, PM averages 93% if it searches each position for one minute. This is on an Athlon 1200 with a hash table with ~3 million slots. PM searches the hash move first (if available). It then looks at positive captures (as identified by a SEE). Then it looks at killers (2 per ply). Then it looks at non-captures which don't appear to chuck away the moving piece; these are ordered by the history heuristic. Then it looks at moves that seem to lose material. After all that, I hope this is what you wanted :-) Andrew
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.