Author: leonid
Date: 09:35:51 08/29/01
Go up one level in this thread
On August 29, 2001 at 10:15:16, Angrim wrote: >On August 28, 2001 at 21:47:33, leonid wrote: > >>On August 28, 2001 at 16:21:30, Angrim wrote: >> >>>On August 28, 2001 at 05:39:13, leonid wrote: >>> >>>>[D]6kq/rqqrqqqq/n2Q4/b1PNRBBN/b2Q3N/p2Q3K/n2P2Q1/q1Q2RR1 w - - >>>> >>>>Please indicate your result. >>>> >>>>Thanks, >>>>Leonid. >>> >>>Athlon 1.2ghz, pn^2 without pn-transpositions: >>>proved that move f5xh7 wins, 11 turns >>>PN2:1146353 evals, 31056 expands, 3.91 seconds >>> >>>Athlon 1.2ghz, pn^2 with pn-transpositions: >>>proved that move f5xh7 wins, 11 turns >>>PN2:952464 evals, 22224 expands, 3.36 seconds >>> >>>Athlon 1.2ghz, pn-search with transpositions: >>>proved that move f5xh7 wins, 11 turns >>>PN:261248 evals, 6348 expands, 24 max ply, 0.95 seconds >> >> >>Hi, Angrim! >> >>What is finally your time for solving this position? It look like it is very >>short but I am not certain what, between few figures that I see here, is final >>number. > >The numbers shown are for three independant searches each useing >a somewhat different method. The fastest time to solve it was 0.95 seconds. >The reason that I do not use pn-search(rather than pn^2 search) for most >of your puzzles is that pn-search is memory bound and so if it can not >solve the problem before I run out of ram then I have to use pn^2 search >which uses vastly less memory at the expense of some(usually 2-3x) speed. 0.95 second for this position is Superb! I just looked into your today response that is also better that mine for selective. Your was somewhere around 20 seconds where mine is 63 seconds. Very good news! I have formidable competition for my positions to make entire sourrounding here very animated and enjoyable. Cheers, Leonid. > >> >>I tried to guess already few time if I have something in my solver like your >>pn-search but never had clear response. >> >>My time was very bad on this position. It took 11 seconds (your look to be 3.36 >>seconds) and selective find mate only in 10 moves. By brute force it took 12 min >>and 55 sec to find shortest mate in 8 moves. Celeron 600Mhz. No hash. >> >>I don't think that hash give that much help on selective search since there are >>only few nodes used on each ply. But hash should be very helpful on brute force. > >The importance of hash tables for selective search is transposition >detection. > >> >>Cheers, >>Leonid. >> >> >> >>>pn-transpositions meaning that the search checks for transpositions in >>>the secondary pn-search as well as in the primary pn2 search. I >>>just recently implemented this for use in my suicide chess program, >>>but it seems to also pay off for standard chess. This costs a bit >>>of speed(nps) but seems to be well worth it. >>> >>>Also clearly raw pn-search is much(3x) better than pn^2 search, but >>>only if you have enough ram(pn-search is best-first search so all nodes >>>must be stored in ram) >>> >> >> >> >> >>>Angrim
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.