Author: James Swafford
Date: 17:42:36 01/20/02
Go up one level in this thread
On January 20, 2002 at 20:08:24, Jon Dart wrote: I always love reading the programmer's post-analysis. Thanks for taking the time Jon. -- James > >CCT-4 notes > >Here are some notes re Arasan's participation in CCT4. > >I had a worry a couple of weeks ago when my wireless network quit >working. I have a couple of Linux boxes and a network hub in the >basement. But the fast machine, the Athlon, is upstairs on a wireless >link. It broke, and it would be a pain to move the Athlon downstairs >onto the wired net, so I was afraid I'd have to use a Pentium III/733 >for the tournament. But I was able to get a replacement wireless card >in time and all was well, hardware-wise, anyway. > >I am still tuning Arasan. In the last year or so I have made king >safety changes, extension changes, bug fixes, and built a new opening >book. Oh yeah, and ported it to Linux. None of this is publically >released yet. Soon, I hope. > >Arasan was paired against Ferret (!) in the first round. Bruce reminded >me that Arasan has played several games against Ferret on ICC under >the handle "Mink", but Mink was on a lot slower hardware. Ferret had >a dual Athlon setup for the tournament. Arasan played a somewhat >passive English Opening setup. Not bad, but not very active. Ferret >crushed it. > >Next round I had PolarChess. Arasan got an early advantage and won >easily. > >Round 3 Arasan was matched with monsoon. Arasan has played monsoon >on ICC. It is a strong engine. I am impressed, in fact, with how >many good amateur engines there are. Chester, Insomniac, monsoon, >amyan, tao, to name just a few. > >monsoon played the Benoni, which I think is just bad for Black. But >Arasan chose the Four Pawns defense (A68), which is far from the >best system. 7. Nf3 is better than f4. I think White can also play g3. >Arasan played into the f4 line because there were some draws and >wins with it from the White side in its book database. But I think >15 .. Nb6 may bust this line. After that, Arasan played one move out >of book (16. Qb3) and the next move was failing low. It doubled the >search time but still had a bad position, and it never really >recovered. Which is not to deny that monsoon played well. But I took >the f4 line out of the book since this game. > >Round 5 today Arasan was matched with Tao. Arasan had the White side >of a Tarrasch defense. At least that's what I think it was. Tao gave >up a Rook for Bishop plus two pawns at move 25. A short time later it >had posted a rook on the 7th and advanced its passed pawn, with a >decisive advantage. > >Round 6, Arasan played Avernox. Arasan has a reasonable record against >Averno on ICC. But here it hit another opening problem, playing the >Pirc defense. After 1. e4 d6 2. d4 Nf6 3. f3 d5 4. e5, Arasan was out >of book and soon had a miserable position. The pawn on e5 continued >to cramp it through much of the game. It was able to exchange down >and escape immediate danger but wound up down material in the endgame >and succumbed. > >This showed up a problem in Arasan's book selection code. Normally it >scores each possible book move and uses a combination of move >frequency, win/loss percentage, and book learning to score each move. >But move selection is then random. It will prefer moves with high >scores, but as long as the score is not zero, the move has some >probability of being selected. So .. d6 after e4 is a possibility, >although not likely. > >I now plan to implement some selectivity. I think Crafty has a similar >feature: if I set the selectivity to 70, for example, it shouldn't play >moves whose score is less than 70% of the highest move's score. This >will narrow the opening book selection but also steer it into more >main lines. I almost have this coded and it should be in place by next >week. I think randomness is good against human players but against >computers you probably want a higher selectivity value. > >Round 7, Arasan played Armageddon, which is a Polish engine I've never >heard of before. Arasan dropped a pawn early in the opening and had to >play most of the game with a minus score, which was not fun to watch. >However, in the end it wound up in an endgame in which the extra pawn >was ineffective. Arasan had only 4-man tablebases and its opponent >had the full 5-man set, so it was lucky to draw here: it was depending >on search while Armageddon was getting steady tablebase hits (according >to the operator's comments during the game). > >Well, it was fun, and I learned some things. So I'm looking forward to >next week.
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.