Author: Tom Kerrigan
Date: 02:03:18 12/29/01
Go up one level in this thread
On December 29, 2001 at 01:24:38, Uri Blass wrote: >Ignoring tactics also does not seem to be a good idea because if there is a good >chance that tactical tricks are going to work then it should be considered in >the evaluation. No. Not unless you have tactical terms in your eval function. Otherwise you will just be changing weights at random. >It can be useful because a team of some hundreds of people may waste only one >year. All coefficient optimization algorithms depend on a way to measure the quality of a set of coefficients. How do you propose to do this? I don't think a test set of any reasonable size can do this. The only other alternative, as I see it, is to play games. It takes on the order of a hundred games to see if one program is better than another, never mind relative performance. And you would want to optimize against several programs, not just one. And you would have to do it at the time controls you wanted to optimize for. You might be talking about running thousands of games at tournament time controls for _one_ single iteration of your weight optimizing algorithm, and while it might be useful to have many computers evaluating different sets of weights in parallel, it's most certainly necessary to do several sequential iterations, so you're talking on the order of several years... >Of course if you change it too much the program may lose games in the way that >you describe but my observation is that practically there are more games when >programs are losing because they put the knight at the corner than the opposite >case. I'm not saying that it's always good to have a knight in the corner. I'm just saying that it's not always bad, and that's the problem. -Tom
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.