Author: Robert Hyatt
Date: 05:42:37 06/03/98
Go up one level in this thread
On June 03, 1998 at 00:51:13, Komputer Korner wrote: > >Okay I will use the KK Kup book from now on, but when I used the small >book that was created from a 400 game Morphy pgn file from M-Chess 7.1, >I used the n parameter as well. that's not really a "book". Because 99% of the moves will not be trusted since they weren't played more than once, and it also won't give much "coverage" for the normal openings either.. >BTW there is a typo in your >documentation on this. The n parameter specifies that unless the move is >played in at least n number of games that move won't make it into the >book. So I created the book this way so that there were more than 1 >result for each move. It pruned the book down to a small amount but >there should definitely have been wins in the book after 1.e4e5 >2.Nf3!!!!!! I saw Crafty out of book after this sequence. later on in >the match it was't out of book until the sequence 1.e4e5 2.Nf3Nc6 >3.Bb5Bc5 4.c3 So the book wasn't corrupted but just that Crafty refused >to use some of it early on in the match. the typo has already been fixed... Tim Mann spotted it a week or two ago and told me... And remember, it is not only *wins* that count, it is the number of times a move is played. With 400 games, after the first 3-4 moves, you are into single played lines pretty much. and they are useless and not trusted by crafty.. >The previous series of Comet- Crafty matches started out okay from a >fairly large book of Kasparov games but eventually after 50 or so games, >Crafty refused to use it's book any more. That was the series where both >programs were pondering and I had crashes all the time in Winboard. >The way you have implemented book learning in Crafty forces users to I have no idea what you mean here. I haven't rebuilt my book in over a year, after 50,000 games played. So that makes no sense. >rebuild books each time to make Crafty's book larger. This is against >some lofty AI principles of having the machine learn by itself. It would >truly be impressive if Crafty could make it's book larger by itself >based on results of its games (MChess 7.1 can do this) and even more >impressive if it could start out from no book at all and build it's book >this way. . I realize that to implement this could mean different >versions for servers and others because of the fast time constraints on >servers, but there are eons more Craftys out in the rest of the cyber >world than there are on servers. I guess it depends on your goals with >Crafty. Based on your results, how many ELO points is Crafty suffering >from playing on a Pentium 166 as compared to a Pentium Pro or beyond. I I don't know about your AI goals, but book learning works perfectly. It was taken and ported directly to Fritz, for example. There's *nothing* in AI that says a book has to be added to.. that wasn't a goal, and it is a bad idea, unless your program runs on a fixed platform all the time. I would not want to run it on a P5/166 and have it add to the book, then run it on a 16 processor alpha and have it use that old added line, because the faster version can see *far* more than the slower one, and it is pointless to trust such analysis. *unless* you design like the commercial programmers do for a user using a fixed platform all the time. My goal was to develop an algorithm that could ingest large numbers of PGN games, and then learn which are good and bad by playing them, without *any* human intervention at all. In that regard, it works perfectly... >don't need the point comparison in relation to speed of machine but just >the extra points that Crafty gets when the 64 bit code is optimized. In >other words Comet against Crafty on a pentium pro as compared to Comet >against Crafty on a Pentium 166Mhz. How many more ELO does Crafty get >this way after factoring out the speed increase from a 166 to a Pentium >Pro 200? >-- no real idea. The issue is null-move search and my simplified quiescence search. As it slows down, it will reach a point where it plays pretty badly making tactical mistakes. We can compare notes after you run a match with a reasonable book, as I now have several dozen games between Crafty and Comet played on ICC, plus dozens more played against *other* crafty versions and comet, on the same servers... >Komputer Korner
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.