Author: Robert Hyatt
Date: 13:48:28 08/19/99
Go up one level in this thread
On August 19, 1999 at 01:25:02, Will Singleton wrote: >On August 18, 1999 at 20:40:33, Scott Gasch wrote: > >>On August 18, 1999 at 19:26:27, Dave Gomboc wrote: >> >>[snip] >>> >>>Tell us the name of your program too! How long have you been working on it? >>> >>>Dave >> >>My program has no name yet... I've been working on it for the past couple of >>months (with help from y'all). It's a PVS w/ null move, killers, history, >>transposition tables on a 0x88 board representation. I am thinking about >>calling it "darkhorse" once it gets better. Right now it is just called "chess" >>:) >> >>I am reasonably satisfied with the search speed and tree size. I've contained >>the massive branching in the qeval by using a SEE. My two major goals in the >>short term are to get extensions working without a big speed sacrifice and to >>improve the evaluation routine. I need to work on an opening book too. >> >>Speaking of eval routines: has anyone ever done any genetic programming to learn >>the values of the eval routine? I was thinking of randomly coming up with eval >>term coeficients and having the computer play itself. Then the winner survives >>and the loser dies... This way I can hone in on "better" eval routines via >>natural selection... anyone done anything like this? >> >>I'll post some samples of the extension explosion when I get home. >> >>Scott > >Let me know when you get it up and running on ICC. Some folks are reticent >about doing that, but it really helps tune the old eval (as well as find bugs). > >I like the name Darkhorse, except that it makes me think of dark horse stuff ;-) >(sorry). > It has also been used previously. I played a program called "dark horse" in one of the middle-70's ACM events... >As for learning eval params, there has been a lot of work in this area. Check >out the TDChess page, or ask Jon Baxter, or look at some ICCAJ's, or check the >links in the resource center here. TD learning seems better suited to >backgammon, but who knows? Anyways, there's a ton of research on the subject, >knock yourself out. > >Will
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.