Author: Sune Fischer
Date: 04:31:17 05/27/04
Go up one level in this thread
On May 26, 2004 at 19:34:48, Dann Corbit wrote: >On May 26, 2004 at 17:55:25, Uri Blass wrote: >[snip] >>people do not expect programs to perform normally in game in 1 second but they >>expect programs to perform normally in game in 1 minute. >> >>bullet is 1 minute per game and not 1 second per game. > >Then they have a bad expectation. Every program makes horrible moves on a slow >machine at game in one minute. > >Even if the machine is fast, most programs will make awful blunders from time to >time at that time control. At 1 minte per game you can typicly search for 1-2 sec per move, that's about 6-8 plies for most engines. Like in the good ol' days :) >The programs that do not crash or even the programs that do not lose on time >overage will still make really bad choices. Doesn't really matter, the engine that wins will still be the one making the fewest mistakes :) In fact it is often easier to see how the engine evaluates when it's not supported by a deep search that clouds the picture. E.g. I found lots of errors where it would play very unsound moves because it just couldn't see deep enough. As a not very strong chess player I really need to weaken it like this to stay on top of things :) These super rapid games are also great for stress testing, if nothing else. >I submit that on a 300 MHz machine, there is no program anywhere that plays well >at G/1 > >At any rate, the slower the machine and the faster the time control, the more >silly and absurd the output. > >But you already knew that, of course. As a developer, an important job is to locate weaknesses and to gather sufficient data. Long games are not very usefull for either. -S.
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.