Author: José Carlos
Date: 01:26:04 08/17/01
Go up one level in this thread
On August 16, 2001 at 20:23:39, Dann Corbit wrote: >On August 16, 2001 at 20:21:55, Peter McKenzie wrote: > >>On August 16, 2001 at 17:56:58, Dann Corbit wrote: >> >>>On August 16, 2001 at 17:51:05, Roy Eassa wrote: >>> >>>>On August 16, 2001 at 15:50:28, Dann Corbit wrote: >>>> >>>>>On August 16, 2001 at 15:42:51, Frank Phillips wrote: >>>>> >>>>>>Mine takes forever and 12 ply to see the trouble, even if I turn up the open >>>>>>h-file with rook (queen) on it king safety term. >>>>>> >>>>>>Do the programs that solve this quickly do it by shear speed, evaluation or >>>>>>clever extensions? >>>>> >>>>>Phalanx solves it very fast. >>>>>Phalanx searches less than 150K NPS on my 950 MHz machine. >>>>>It has very clever search extensions. >>>>>It has very clever king saftety. >>>>> >>>>>So darn clever (in fact) I have a hard time figuring out what is going on, >>>>>despite having the source code at my disposal. >>>> >>>> >>>>Phalanx, eh? How has it done in tournaments vs. the other engines? If not so >>>>good, what weakness(es) does it have to offset these strengths? >>> >>>It finished about the center of the pile in "The Battle of the Crowns" chess >>>tournament in the 4 crown division. (This was over a year ago, and many of the >>>program versions are now outdated). >>> >>> Program Score % Av.Op. Elo + - Draws >>> 1 LGoliath : 37.0/ 52 71.2 2289 2445 76 99 26.9 % >>> 2 Crafty : 32.0/ 52 61.5 2294 2376 85 77 34.6 % >>> 3 AnMon : 31.5/ 52 60.6 2294 2369 86 83 28.8 % >>> 4 Amy : 28.5/ 52 54.8 2297 2331 93 85 21.2 % >>> 5 SOS : 28.0/ 52 53.8 2298 2325 94 82 23.1 % >>> 6 Bringer : 27.5/ 52 52.9 2298 2318 96 72 32.7 % >>> 7 TCBishop : 27.0/ 52 51.9 2299 2312 97 73 30.8 % >>> 8 Phalanx : 26.5/ 52 51.0 2299 2306 99 74 28.8 % >>> 9 Comet : 24.5/ 52 47.1 2301 2281 62 96 44.2 % >>> 10 Gromit3 : 24.5/ 52 47.1 2301 2281 79 96 25.0 % >>> 11 Gromit2 : 21.5/ 52 41.3 2304 2243 89 88 21.2 % >>> 12 Francesca : 18.5/ 52 35.6 2307 2204 102 82 17.3 % >>> 13 Yace : 18.5/ 52 35.6 2307 2204 87 82 28.8 % >>> 14 ZChess : 18.5/ 52 35.6 2307 2204 108 82 13.5 % >>> >>>The engine is 150K NPS on a very fast machine. Other engines are much faster >>>and get one or two full plies deeper at every turn. >>> >>>Attach a really fast motor, and Phalanx might rule the world. >> >>If it had a really fast motor, maybe it wouldn't have time to do the fancy >>extensions and fancy king safety stuff? > >Well, good point. > >I really don't know where the brakes are in the system, since I have not done >careful profiles. > >But I suspect the move generator. Most engines develop 3-4x the NPS of Phalanx >on my machine. Maybe the darned clever search and all the fancy eval stuff is >the reason for that. But I ?suspect? not. Last time I had a look at Phalanx code I saw something that can explain it's slowness: When it believes it's near the endgame, it does both evals midgame and endgame, and then chooses a in-between value, depending on how close to the ending the game is. I haven't studied its move generator, but it's usually accepted that the move-generating time is not more than 20% of total search time. So having a slow move generator shouldn't make a program 'dead-slow' itself. José C.
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.