Author: Uri Blass
Date: 16:07:20 10/06/01
Go up one level in this thread
On October 06, 2001 at 18:58:33, Dann Corbit wrote: >On October 06, 2001 at 18:19:49, Slater Wold wrote: >[snip] >>I just think that you should respect a mainline, more than an eval. Evals are >>very volatile, while mainlines are more solid. >> >>Once again, Program-A sees a position as winning, and gives it a +6.00 score. >>Program-B sees the _SAME_ exact moves, and gives it +3.00. >> >>Which is correct? The one with the bigger eval? I just don't think that can be >>trusted. > >Mainlines are just as volatile as the score. While I trust neither, I trust the >score better than the mainline. Ideally, it will have both the score and the >mainline correct, and then we can feel more secure about it. The main problem with scores is that +3 for one program may mean the same as +4 for another program. The score is not something that has the same meaning for all the programs. big score can also be result of a bug in the evaluation when the chance that it happens with correct main line is smaller. > >We must ask ourselves this question: >If the mainline is correct, then why is the score wrong? The only answer can be >that the computer has not found out yet why the move is good. It may very well >be that there is some tiny tactical bonus found by taking the correct mainline, >but the actual purpose is positional. Not only will the computer never get the >true appraisal, but when it gets near the goal, it will play stupidly because it >never does find the plan behind the move. If it finds the correct main line then it usually means that it is going to find the right moves. problems when the program needs to find also something that is positional many plies after the root position are rare. Uri
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.