Author: Robert Hyatt
Date: 19:43:45 05/21/04
Go up one level in this thread
On May 21, 2004 at 16:44:38, Bryan Hofmann wrote: >On May 21, 2004 at 16:10:25, Robert Hyatt wrote: > >>On May 21, 2004 at 15:38:45, Matthew Hull wrote: >> >>>On May 21, 2004 at 15:17:55, Robert Hyatt wrote: >>> >>>>On May 21, 2004 at 15:05:05, Chris Taylor wrote: >>>> >>>>>http://www.chessbase.com/download/engines/crafty/crafty1913.zip >>>>> >>>>> >>>>>Sort of say it all! Thank you Bob! For providing Crafty for so many people!! >>>>> >>>>>Chris >>>> >>>> >>>>With what Anthony is working on, we might actually see a significant strength >>>>improvement with automatic evaluation tuning. I've spent the last week making >>>>all the eval terms modifiable on-the-fly for his code. Also now we can really >>>>start to do some "personality development" as _everything_ can be changed and >>>>saved to files... >>> >>> >>>Did Anthony use this process to tune Zappa? It's strength was impressive at >>>CCT6. >> >> >>I don't know but I don't believe so. But obviously we are going to apply it to >>both programs. It is producing some interesting "values" already but there are >>more changes to be done. And it takes a bunch of compute cycles. He's working >>on a distributed implementation to run on our 32-node dual-opteron cluster, >>which will theoretically arrive in 14 days or so (current status says it is "in >>build" so it is getting close. >> >>I'll leave further explanation to him since it was his idea... > >Something simular to this? > > http://www.cs.ualberta.ca/~tony/RecentPapers/jcis01.pdf maybe some slight similarity. The real problem is that Crafty has something like 120 different eval terms, many of which are large arrays of data. Trying to tune _all_ of that is a big undertaking. Here we are really talking about tuning eval scores however, rather than trying to tune search controls...
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.