Author: Robert Hyatt
Date: 17:45:00 01/04/98
Go up one level in this thread
I remain convinced that evaluation and search depth go hand-in-hand. I got burned badly once by not paying attention to this issue, when I really already knew about it. The story goes like this: In 1984 Cray Blitz won the ACM event, and the final result was probably the most convincing victory we ever had, because we beat the programs that finished in 2nd, third and fourth place, a pretty difficult pairing (these were the Spracklen's program, BeBe and NuChess). The next Summer I was getting ready to move to Birmingham to start work on my Ph.D... Bert and I played many games that Summer using the Vax 11/780 for Cray Blitz, and we used both Chess Challenger and a SuperConstellation as the opponents. We were searching 4-5 plies on the vax for reference. We were winning more than we lost (the vax was only 4+ plies slower than the Cray to put this in perspective) and were fairly happy, but we kept noticing a tendency for CB to push pawns and create holes that would later "haunt" it in the endgame. I added some code to penalize holes (took a whole 4 lines of code since we were designed around "that" vector machine.) CB played better (using the VAX for testing) and we decided we liked the more "Karpov-like" style it was playing. I moved to Birmingham, and in October Bert and I went to Denver for the 1985 ACM event, where "HiTech" was unveiled for the first time. We ended up losing two games and winning two, and I simply wrote it off to "they are catching up to our speed" and didn't study the games. Big mistake. We then went to Cologne Germany for the 1986 WCCC and won easily in round 1, but in round 2 we lost to Bobby. And we lost badly thinking we were winning. In this game, we castled long, and at one point had our pawns on a2, b2 and c2, while Bobby had his on a4, b4 and c4. We thought we were winning, counting all the pawn holes black had behind those pawns he had advanced. As a result, we got crushed. After that game, we finally figured out something was wrong, and spent the next 8 hours smoking a Cray looking for answers. On a whim, I deleted the 4 lines for pawn holes, and Viola! Cray Blitz "came back". We tried the Bobby game and it immediately found and played the right moves to prevent that pawn-storm. We played other games we lost and in each game it played much better moves. We decided to stick with this "new and improved" version without the pawn hole stuff, and if you look at our round 3 game vs BeBe, you'll see a computer (CB) completely roll BeBe into a small ball with a nice king-side attack, like Cray Blitz of days gone by. We also won round 4, finding a very deep (17+ plies) win in an ending that was nearly a draw, but using Cray Blitz, Mike Valvo found that its analysis was correct and it was, in fact, winning easily. In round 5 we also whacked HiTech and won our second straight WCCC title. The moral? The pawn hole code definitely improved CB at shallow depths. But CB understood outposts and weak pawns and so forth, and going from 4-5 plies to 9-10 plies made that pawn-hold code simply a "double penalty" because it got a penalty for the hole, plus it got a penalty when it saw how to utilize that hole. I've been careful since, although on ICC this is a problem since we play game in one minute as well as game in one hour or more... :)
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.