Author: Tim Foden
Date: 01:55:52 09/25/03
Go up one level in this thread
On September 25, 2003 at 03:24:27, Johan de Koning wrote: >On September 24, 2003 at 03:34:14, Tim Foden wrote: >>On September 24, 2003 at 01:58:35, Johan de Koning wrote: > >Well, I'm not going to hack more code to create a game playing loop. >It was a bit stupid of me to create Kingless basics. All that was needed was >paralyzing the Kings and creating a PG specific search, leaving the interface >and all the utility code unchanged. Yes, this is what I did in GLC. >But more importantly, it seems pointless to play this game on an 8x8 board. True enough :) >The initial position will be solved tomorrow, and a small opening book will >suffice to win all games. That's assuming White does win, based on the seqence >sofar: =-=+=+= for symmetrical baseline positions 1 through 7. It does seem likely to be the case. >>On the 7-7 position, I decided that the lack of left/right symmetry was hurting, >>so I stopped that too. :) > >It is perfectly symmetrical if you disregard the a-file. :-) Yes, but I couldn't be bothered to change my hash symmetry code to do it. :) >But I don't think symmetry gains much, since the root moves in the "other half" >spawn small subtrees. Maybe it helps a lot if you apply *all* translations and >mirrors to the transposition table. Yes, this is what I've done. I generate 2 hash keys (horizontally reflected), and load/store in the transposition table using the lowest valued key. In my tests it made nearly a 2x speedup. I've also tried with 4 hash keys, including black/white symmetry, but I got strange results, which were only marginally better in some cases, but mostly worse. I've yet to figure out if this was a bug in the implementation, or a flaw in the theory that this would be a correct thing to do. >But that seems to be a bit too much for a >one-time fun experiment. :) >364.77 420M977 D25 +5 : d4,d5 h4,f5 h5,f4 g4,fxg3 fxg3,g5 hxg6,hxg6 g4, ... >734.37 849M691 D27 +0 : d4,f5 c4,g5 e3,h5 f3,e6 d5,e5 g3,f4 gxf4,exf4 e ... >1015.8 1172M D27 +5 : f4,d5 b4,f5 b5,d4 h3,c5 bxc6,bxc6 c3,dxc3 dxc3, ... >1267.2 1466M D29 +0 : f4,d5 b4,f5 b5,d4 h3,c6 bxc6,bxc6 g4,g5 fxg5,fx ... > etc >2207.3 2577M D45 +0 : f4,d5 b4,f5 b5,d4 h3,c6 bxc6,bxc6 g4,g5 fxg5,fx ... >2273.4 2655M D45 14/14 : 1168.0 kN/s >0000011111 1111111222 2222222222 2222111111 00000::::: :::::::::: :::::::::: >0347801335 5668899012 2333343333 2200885532 99552::::: :::::::::: :::::::::: > >Here it stops because the deepest nodes were 44 ply, hence there is no more >information to be gained by searching deeper. (I'm not sure if this theory >holds with transpositions from higher drafts, but my intuition says it does.) Fair enough. :) >[D]8/ppp1pppp/8/3p4/3P4/8/PPP1PPPP/8 w - - > >>I thought of trying the 8-8 position after d4 d5, but I forgot to leave it >>running last night. Maybe I'll try it tonight. > >You already posted a Mate16 for this one, Hmm.. I completely forgot that :) >though that was with stale-ep. >Anyway, JJ agrees with Mate16. > >015.76 18M151 D19 +5 : e4,dxe4 h4,e3 fxe3,e5 dxe5,c5 e6,fxe6 h5,c4 h6, ... >055.69 64M585 D21 -9974 : e4,dxe4 h4,h5 d5,c5 dxc6,bxc6 b4,f5 g3,e5 b5,cx ... >093.54 106M207 D21 +5 : f4,f5 a4,a5 c3,b5 axb5,a4 h3,h6 b6,cxb6 g4,fxg4 ... >191.03 219M386 D23 +5 : f4,f5 a4,a5 c3,b5 axb5,a4 h3,h6 b6,cxb6 g4,fxg4 ... >412.22 481M983 D25 +5 : f4,f5 a4,a5 b3,c6 b4,axb4 a5,h6 h3,b3 cxb3,g5 f ... >745.39 871M353 D27 +5 : f4,f5 b3,a5 c4,dxc4 bxc4,a4 d5,a3 h3,c6 dxc6,bx ... >983.17 1149M D27 +9969 : g4,g5 a4,a5 c3,e6 e3,c6 b4,b6 e4,dxe4 bxa5,bxa5 ... >1113.3 1305M D29 +9969 : g4,g5 a4,a5 c3,e6 e3,c6 b4,b6 e4,dxe4 bxa5,bxa5 ... >1366.7 1608M D31 +9969 : g4,g5 a4,a5 c3,e6 e3,c6 b4,b6 e4,dxe4 bxa5,bxa5 ... >1461.0 1723M D31 14/14 : 1179.4 kN/s >0000010111 1111111112 2222222222 00:::::::: :::::::::: :::::::::: :::::::::: >0347609214 3647587980 0213243424 43:::::::: :::::::::: :::::::::: :::::::::: > >An (final?) improvement I added today is sorting "the reamining moves" by a >fixed advance/center table. The advantage is that the order no longer relies on >on the evaluation through killers. The evaluation can now be skipped without >measurable penalty, resulting in more predictable search. Now using a fixed >eval of +5 for White, the root only changes when disaster strikes, as can be >seen in the above PVs. > >For the record, the outputs above come from an XP2200(1800MHz) with a 16M entry >TT (with experimental replacement scheme). Since the node counts peak in their >late 20s, I'm expecting 27 or 29 to be the heaviest iteration on the initial >8-8 position. It's running already and has produced a 27 ply PV in less then >2 hours. Full report will follow. I'll look out for it. I've stopped working on it for the moment, as real work is interfering. :) Cheers, Tim.
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.