Author: Shep
Date: 02:30:15 10/22/99
Go up one level in this thread
On October 22, 1999 at 03:06:37, Martin Grabriel wrote: >Below is just my pure fantasy (no malice intended, just thinking out loud): > >Somewhere in the near future.... >The brain of any chess program is its evaluation function. What if CM10k comes >with a learning function that could adjust its evaluation parameters on its own >to get 98% to 100% match of moves in a database of games of any of the top-of- >the-line programs. Then it would mean that it could clone itself to the program >just like what Charles is doing manually. Imagine creating a database of Hiarcs >10.64 (64 bit:)) games and let CM10k analyse through (with auto-adjusted >analysis time per move to mimic hardware and time control used in each of the >actual game in the database). Then CM10k stores the evaluation parameters as a >`Hiarcs.cmx' personality. That would make you want to buy CM10k ? Then >Mindscape may want to pay Charles for the idea. > >CM's analysis at present already measures in % terms the moves made by a >player/another program to which it agrees or disagrees based on CM's evaluation >funtion. It could easily be programmed to work backwards to get close to 100% >agreement by adjusting its evaluation function. The end result is that >Hiarcs.cmx might not be identical to the real Hiarcs, but as a `clone' it would >have resemblance. > >In fact not just CM, any of the existing programs could have this new cloning >function for its next function. I think if it were that easy, we would have a Super-GM program already. Just take the 600+ Kasparov games from most common databases and auto-adjust your program's parameters to it... But it's like test-suite tuning, it generally won't make the program play much better. --- Shep
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.