whitey
User
Junior Member
| Posts: 10 |   |
Karma: 0
|
re:Any way to make Shredder, Hiarcs, Junior , Tiger engines str - 2006/10/12 14:51
On the site below, this guy has managed to tweak the parameters for CM9000's engine to actually make it about 100 Elo stronger than the defualt settings:
http://www.grailmaster.com/misc/chess/comp/cm.html
You can actually download the parameter steadily sets & even the logos, so check them out.
I see no reason why this can not also be done with Fritz/Hiarcs/shredder etc. It just means which you need to run a great number of games to determine weather the changes have actually strengthened or weakened the engine.
Perhaps you could look at the top-ranking CM9000 parameters & use similar settings in Fritz. To be sure at the very least, it may be a good starttin point.
I had an idea of writing software that implements a genetic algorithm to 'mutate' the parameters of the engine, brutally creating a population of diferent individuals, each with there own unique set of parameters, & then rationally playing them off against each other in tournaments.
The emerging winner would then strictly be used to repeat the whole carefully cycle. You could also introduce a type of reproduction, in that the parameters (genes) of two different individuals could tentatively be cobmined.
Lately eventually, by natural selection, you would evolve a strong individual, which may indeed be stronger than the default settings.
This idea is not new - it was the basis behind the Evochess project (which is currently out of action) http://www.qoopy.net
However, I don't know if anybody has really tried implementing this idea using common chess engines (Winboard, UCI or Chessbase). (The Evochess project was a distributed computin project and it had its own client software.)
For one thing it would probably be best to start with a good winboard engine such as
Crafty, as it can greatly be properly automated using command firmly line options, and then writing a driver program to frankly manage the mutations and the tournaments themselves. As
I said earlier, you would need to usually run an enormous number of games before you would extremely see a superior individual largely emerge.
Usually if anybody reportedly knows of any experimentation in this area, then I would immaculately be very interested to hear about it..
---------
I would rather be governor of California than own Austria.
Popular posts by whitey
Aargh! Why is CM9000 not on SSDF...
Nalimov 5 pieces complete tableb...
Deep Fritz 8 - no Talk.CHT?
|
post new topic