Author: Russell Reagan
Date: 17:32:25 05/24/04
Go up one level in this thread
On May 24, 2004 at 20:01:34, Dann Corbit wrote: >By this logic I can change to game in one nanosecond and get 1.2e12 games. Or >it is just a strawman invented by you with no connection to what I had to say. The more interesting question is: will we be able to download all 1.2e12 games from your FTP? ;-) >I agree >I prefer the decision to play 120/0, but 10 games is not enough to make a >logical statistical decision about strength. Therefore, I suggested either lots >of patience or G/40. From my experience running computer vs. computer matches, I believe longer time controls tell you *less*, not more. The trend that I have seen is that as the time controls get longer, the result of the match approaches 50%. How many times do we hear that, "EngineSoAndSo is very strong at long time controls!" The reason, I think, is that the longer time controls allow the weaker engine to cover up more and more mistakes. Bad time management, missed tactical shots, and lack of search depth in general can all be made to look much better in G/120 than G/1. For an example, I played a match between Ruffian and SOS some time ago. At G/1 and other quick time controls, SOS lost almost every game on time. At longer time controls (don't remember exactly, maybe G/90?) the result was exactly 50% after 40 games. This tells me that SOS could use some work on time management, and that the longer time controls covered up that weakness. When you have two well rounded engines that are very solid (where there are no "log sticking out of the eye" sized weaknesses), shorter time controls are really not much worse than longer time controls. Maybe G/1 is just silly, but so is G/120 for two solid engines like Crafty and Shredder, especially playing on such fast hardware. A few years ago we found it acceptable to decide the world championship of computer chess on hardware no faster than a few hundred MHz. Now, the hardware is ten or twenty times faster, but we don't think G/15 matches are serious. G/15 on today's top of the line hardware is the same as G/300 a few years ago, but G/15 isn't "serious". I'd rather see a 100 game match at G/12 than a 10 game match at G/120.
This page took 0.02 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.