Author: Robert Hyatt
Date: 10:07:27 02/02/02
Go up one level in this thread
On February 01, 2002 at 09:46:34, Uri Blass wrote: > >This make more sense > >I can beleive that they did 12 ply brute force+extension + >selective search of 6 plies. > >selective search is done by all chess programs and it is called qsearch. > >I can believe that the software did the brute force search when the hardware did >some kind of qsearch. > >I can also believe that their kind of qsearch included more nodes than the >normal qsearch of chess programs and they used usually 100 nodes for doing one >qsearch at depth 6 but it is still some kind of qsearch. I don't know much about their q-search other than the rather mysterious reference to "futility pruning in the chess chip." That could be a futility pruning algorithm applied in the q-search as I do in Crafty (Hsu and I had a long discussion about this in the early 1990's in r.g.c.c) or it could be a full-blown futility pruning approach applied to the last N plies of the full-search as well as the q-search. I won't try to guess without more information... > > > But I said N when to be consistent with programs >>like Tiger of today, I could have said X+4. IE Tiger doesn't do full-width >>searching to depth 16. So Christophe has to explain what that 16 means when >>compared to others. Hsu did the same for Deep Blue... I would agree... particularly when we know that the software search and hardware search are greatly different. And when we really don't know a lot about the last version of DB (DB2) as that introduced the "futility pruning" Hsu mentioned and there is no telling what _else_ they tried and haven't reported on yet... In the case of Cray Blitz, a 9 ply search is very "unclear". Because if you look at any of the papers on Cray Blitz, the search was divided into three "regions". For a 9 ply search, the normal search rules applied for the first 9 plies, including the typical extensions like check, SE and so forth. After that search had died down to 0 plies left (which could have been a 20-30 ply variation including extensions of course) we then tacked on a "threat" search of 4 plies. Here we did all the normal q-search moves, plus we could include other "interesting" moves such as those moving the queen close to the opponent king, or captures near the opponent king, and so forth. Once those 4 plies were done (and they _could_ be extended also so that this region of the search was simply 4 plies minimum, but with the usual extensions (no SE here, just checks, captures, etc) it could be more than 4 plies of stuff, we did the normal q-search, which in the case of Cray Blitz, could include all captures and all checking moves (so long as there was no stand-pat opportunity prior to including more checks since the mate would not be forced). It is hard to say "our 9 ply search" was like a normal 9 ply search today. It was significantly deeper. But we always said "9 plies" and when others compared to us they often said "hey, I am on a micro and I am going 9 plies, what is wrong with that fast hardware of yours?" When we smashed them tactically they would say "oh, I guess your 9 plies is not the same as our 9 plies..." :) comparing a simple term like "ply" is therefore very misleading... > >It is known that christophe does not do brute force search. >The claim of Ed was that Deep blue did not 18 plies brute force search and this >claim is proved. I disagree with that. How can you +prove+ anything? You don't have access to the hardware, nor the software, nor anything but the game logs... To prove something certainly requires some data and we have too little of that at present. > >I also believe that their selective search in the last plies was clearly more >selective than 6 plies of search of all top programs but less selective than the >normal qsearch of top programs. Speculation is fine. I have no idea. And I really don't have any info to speculate on other than the minor note made by Hsu about "futility pruning in the chess chips." > >Uri
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.