Suicide Checkers II

After Suicidal Cake's clear victory against Roshi47 (see my match report) I was pretty confident that Suicidal Cake was a great suicide checkers program - was I ever wrong! A few days ago, I received the following email:

From: Igor Korshunov
To: Martin Fierz
Subject: suicide checkers chalenge
Date: Fri, 21 Apr 2006 10:14:01 +0300

Hi Martin!
I am made first version of suicide checkers.
I want to match your program.
When we can meet in battle?

I have 2 propousals:
- on-line match
- play by e-mail, with 7 first possible moves

Best wishes,

Igor is a 28-year old programmer from Belarus. He wrote a strong chess program (WildCat) and also a strong Russian checkers program called Kallisto. His SuicideKallisto ran on a Sempron 2520MHz with 2GB RAM which is probably very slightly faster than the Athlon 64 3000+ I was using. SuicideKallisto (SK from now on) only had the 6-piece endgame database and no opening book - he would have no chance at all, I thought.

The Match

On April 25, 2006, the match was played on Kurnik. Once again it turned out to be a more lopsided affair than I had expected. Suicidal Cake (SC) lost all 4 games we played. Game 1 seemed to be petering out to a draw with both SC and SK seeing draw scores, but finally it turned out that king centralization is very important in suicide checkers endgames, something that SC didn't know and it duly lost what appeared to have been a drawn position. Game 2 was even worse: SC got a high book score, grabbed material and went down in flames. Game 3 was another hopeless affair, with SC grabbing material and losing. Finally in game 4, a transposition of moves got us back into game 2 after only 3 moves, and I resigned immediately - the programs would just have repeated game 2. Igor generously offered to forget about this game, and play a different opening, but I had seen enough and was quite disgusted with my program's performance. It was clear that SK was playing in a different league, with a far deeper understanding of the game than SC's simple materialism. Check out the games below to see what I mean. Congratulations to Igor for a fine program and a clear victory! I thought my larger database and the opening book would give me a clear advantage; but as it turns out, my opening book is useless: a book that is computed by a weak engine is just a huge pile of mistakes, and I'll first have to improve SC before computing a new book.

If anybody reading this spots a general rule about suicide checkers that SC was violating and that SK understood, please tell me what it is by sending a mail to nospam1 at fierz dot ch - thanks! Of course, I'm trying to find out myself too, but any help is appreciated :-)

The Games

The games have Cake's evaluation as comments
-- April 26, 2006