Author: martin fierz
Date: 14:33:13 05/21/03
Go up one level in this thread
On May 21, 2003 at 17:09:00, Tom Kerrigan wrote: >On May 21, 2003 at 15:51:13, martin fierz wrote: > >>On May 21, 2003 at 11:58:55, Tom Kerrigan wrote: >> >>>On May 21, 2003 at 09:07:24, martin fierz wrote: >> >>>>connect 4 can be solved on a fast pc in 1 hour. checkers is nowhere near being >>> >>>Really? I didn't know that. Not by "brute force," i.e., only scoring positions >>>as win, lose, or draw. I believe that takes closer to a month. >> >>dead wrong! my connect 4 program is pure brute force with a large hashtable. 1 >>hour on a fast PC is enough for it to solve the game. it has an evaluation >>function, but in fact, it also has a search win mode, where the eval is turned >>off and it is a pure brute force searcher, that's faster in most cases! some >>other programs are even faster, maybe they have more intelligence :-) > >Wow. I wonder why it's so much faster than my program. How many NPS does it >search? I think my program searches 2M NPS on my AXP 2000+. All I do is >alpha-beta with a fairly large hash table... mine does a bit more than 1MN/s on my P4 1.4GHz. so i guess they're similar in speed. you will have to define "fairly large" though. i'm talking 512MB here :-) hashtable size makes a big difference for me, because i have very little intelligence in the move ordering - the only good part of MO is the hashtable, and once that's full it gets much worse. two BTWs: -there is a german program called mustrum, which says it solves connect 4 in an hour on it's homepage: http://www.lbremer.de/mustrum_e.html. i know this is pure brute force too. -james allen's program needed 6 billion positions to solve the game in 1990, as i read in that paper by victor allis referenced in this thread. with your speed that would be 3000 seconds - right on the spot :-) >Right, I never said it was solved. But that's an indication of how much easier >checkers is than chess. yes and no. it's not hard to write a very good checkers program if you have an 8-piece endgame database. but it's quite hard to make your program better than the others in checkers - much harder than in chess, because there is very little room between the opening and the endgame where your heuristics can make the difference. but i'll admit: i wanted to write a chess program in fact, and thought i'd first have a go at checkers programming, as the rules there are much simpler - so you're up and running much faster. that was 7 years ago - i got stuck :-) cheers martin
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.