Author: Robert Hyatt
Date: 08:37:56 04/30/05
Go up one level in this thread
On April 30, 2005 at 01:48:59, Steven Edwards wrote: >Embarrassing moments in computer chess > >Every author enjoys the experience of seeing his program shine in an OTB >computer chess event. A clever win, a hard fought draw against a stronger >opponent, and the occasional brilliant move are things that are remembered for >years thereafter. > >But what about the incidents that an author would have rather had not happen? >Not so much fun, yet they are often more instructive. > >I will start out with a story of an unfortunate event that occurred in the OTB >play of my old program Spector. > >Back around 1987 when Spector was less than a year old and was still a K&R C >program, I was in the habit of entering it into some New England tournaments >where it was the only computer program participant. In one of these four round >Saturday swiss events, Spector was running on a Macintosh Plus (8 MHz MC68000, >later upgraded to a 16 MHz MC68030) and started out the first round looking >okay. Well, okay until several moves out of the book. After that the program >started to very gradually slow down as the game progressed. (The node frequency >was displayed along with the PV trace on the Mac's nine inch porthole CRT.) It >lost the first game and the second as well with an even greater slowing down. >After it lost the third game, I started to become suspicious. All doubt was >gone that there was a yet undiagnosed problem after Spector lost the fourth and >last game with its node frequency steadily dropping until it was something like >five percent of what it was in pre-event testing. > >What had happened? I had earlier come up with a "clever" way of recycling move >vector storage. And it worked rather well in the many, many pre-event test >suite runs. (This was pre-EPD and all the suites were fairly short.) Alas, I >had completely neglected to run a single whole game test after my clever >modification, and that modification in practice worked reliably only once per >program initialization. After a game was started, the program never released >much of its allocated storage and so it had to work more and more to allocate >less and less. The four straight losses were an amusement to the human entrants >and a lesson to me to do full game testing prior to an event. > >What other embarrassing moments can be shared? Entered a new version of Cray Blitz in a labor-day-weekend event in Mobile Alabama in 1984. This was a brand new version of the parallel search, based on "PVS" (not the PVS serial programs use but the PVS algorithm as defined by Marlsand and Newborn). We were running on a new 4 cpu Cray XMP, and we were entered because we were testing to get ready for the 1984 ACM event in Los Angeles. We reached a position against a 2500 player, and out of the clear Cray Blitz announced Bxp+ mate. Opponent calmly captured the bishop to get out of mate. The bug? There were three legal ways out of the check. One processor searched a tree of size zero and since the opponent had no legal moves for that processor to search, it concluded "checkmate". :) This was obviously the best score to be found and bang... Being a bishop down we lost easily...
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.