Author: Danniel Corbit
Date: 17:15:50 05/11/98
Go up one level in this thread
On May 11, 1998 at 13:15:16, Don Dailey wrote: [snip] >Your response was well thought out, in my opinion you overestimate >the tuning aspect, although the DB team itself does claim some work >went into the specific tunning. > >By the way, I think reasonable tunning for matches against humans >in general is possible and I have done a little of it myself. I >have no real sense of whether it helped or not. There is some >anecdotal evidence that certain programs are better against people >than others programs (relative to each other.) I have my doubts >the effect is large but I believe it probably exists. This is the evil User923005 at work. Here is how I would tune for an opponent. Take every game that opponent has ever played, as PGN. Sort into each UNIQUE FEN position. Use one computer per FEN, analyzing at 24 hrs per move. Store each analyzed FEN in a database. When you play that opponent, if any position that GM has ever played comes up, you will know immediately [analyzed to far more plies than you can in an actual game because it got 24 hrs compute time] the value of the position, as well as perhaps a kibits of the future. This may sound farfetched, since there are surely thousands of unique FEN's. But (for instance) the organization I work for has hundreds of computers. I simply run a java applet on them when they are idle (for instance, at night). After a few months, I have a database that represents enormous compute power. Whenever I am at a stored position, I am incredibly powerful. But, you may reason, the GM will expect this behavior and play an opening or varyation that they don't usually play. Good, then they will be out of their comfort zone and their element. That is what Kasparov tried, and he got spanked when he did. If we have a database of, say half a million FEN positions that came up in real GM games, and have them all analyzed to 17 or more plies, we will have a distinct advantage when ever we hit that position. The computer will not be able to churn through that many during the game, but it won't have to. Even if only some fraction of the moves during the actual game have been analyzed, whenever we hit one, we will suddenly become brilliant. We can also look for "nearby" FEN positions that are desirable for us or undesirable for the opponent and try to figure out how to get there. We already know their value.
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.