Author: Eric Oldre
Date: 07:33:32 04/13/04
Go up one level in this thread
Thanks Olivier! For you experienced authors who want a trip down memory lane, I can tell you a story of my night last night. "just before bed" (ya right!) I decided to test my engine against the program it would be playing In Olivier's tournament today. It started off good enough, and eventually my engine found itself with queen, rook, knight, and 2 pawns against the opponent's king. It saw mate in 2 but wouldn't play the appropriate move!!! It continued chasing the king around the board until the 50 move rule applied. not good! Eventually I discovered the problem was coming from null-move cutoffs. I had logic to not use null move in the endgame, but the function that detected the stage of the game still said it was late-middle because i had too much material left. I fixed it by saying to figure the game stage by only factoring in the opponent who has less material. so if black is down to just a king, it's the endgame, regardless of what white has. and null moves will be eliminated. (is this how others do the same?) Since i had been up so long, I decided to "play one last test game" (ya right again). This time my program was up a minor piece, and decided to put the opponent in a continuous check... draw by 3-fold repetition, the first thing my search did was check for draws by the 50 move rule and 3 fold repetition, then return the contempt factor if detected. so why didn't my engine see this. I changed when it would use the transposition table, to not use it for anything other than move ordering at the root node. (seems so simple doesn't it) Anyway, I ended up being up way to late, thus the hung over comments, and sent Olivier the updated binary which i hope didn't expose any new bugs, but at this point i'm simply going to cross my fingers. Hope you enjoyed the story. Eric On April 13, 2004 at 03:08:06, Olivier Deville wrote: >Eric, I am not able to send e-mails right now (reception seems to work), so I >wanted you to know that I got last version of Murderhole (1.0.6). > >The game vs Chad's Chess should be played tonight (French time) so if you find >another major bug you can still send me a new version :) > >Olivier
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.