Author: Dann Corbit
Date: 18:01:24 01/13/01
Go up one level in this thread
On January 13, 2001 at 20:56:26, Robert Hyatt wrote: >On January 13, 2001 at 17:39:14, Gian-Carlo Pascutto wrote: > >>On January 13, 2001 at 17:19:19, Robert Hyatt wrote: >> >>>Do as I did. Make the extension amount something you can set via command. >>>Then run a potload of tests. I ran WAC with all the extensions set to >>>values between .5 and 1.0, in increments of .25. That is 3 cases for >>>each extension and I varied 4 different extensions. 81 tests and you then >>>look at which ones needed the fewest total nodes to solve _all_ the test >>>positions... >> >>Doesn't that have the disadvantage of tuning the extensions only >>for WAC, but not for general chess play? You can hardly say that >>WAC are 'typical' positions... >> >>Since I'm not solving all of WAC yet in 'acceptable' time your >>method seems hard to apply. I'm not sure just picking the one >>that just solves most is the right thing to do for the above reason. >> >>Especially because that happens to be the one _without_ fractional and >>recapture extensions. >> >>-- >>GCP > > >Then pick any test set you want. But the point is that you _must_ be able >to reproduce the same test for different extension values to be able to compare >them... > >If you just watch it play games, then you might as well pick random numbers >for the extension values.. I have a suggestion (assuming well known and thoroughly debugged test sets like WAC)... If one program version gets more correct, it should win that run. But if they get the same number of problems correct, then perform a summation of the centipawn evaluation into a double. All the games have some key move that should dramatically increase the eval. The total sum of points should really be telling to say how far along the program is at knowing which moves are good. In other words, version a picks axg5 with a ce of +100 and version be gets a ce of +125. The second version is probably more accurate. In WAC, I think probably all the moves are either winning moves or moves that salvage a draw from certain defeat.
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.