Author: Robert Hyatt
Date: 11:08:42 11/03/97
Go up one level in this thread
On November 03, 1997 at 12:07:54, Chris Whittington wrote: > >On November 03, 1997 at 11:31:37, Chris Carson wrote: > >>On November 03, 1997 at 10:42:37, Howard Exner wrote: >> >>>I think one disadvantage Crafty may have is that it is >>>so easily available to everyone. There is no mystery about >>>its playing style, strengths and weaknesses. >>> >>>Here are some observations to support this theory. Note just a few >>>recent examples of programmer secrecy such as in Deep Blue for one and >>>the withholding of games from the French Championship a few >>>weeks ago as another. Secrecy in chess is commonplace for all chess >>>players. Also,I wonder how many programmers have now included opening >>>preperation against Crafty's ingenious forth move in the Ruy Lopez >>>(B-c4 instead of the common B-a4) that was seen about a year ago? >>> >>>How might Crafty fare in two years from now if Bob took a 2 year >>>break from releasing Crafty to the masses? >>>Call it "Hyatt's Hiatus" (say that quickly 3 times in a row). >> >>I agree with you and with Chris W. (in a different post). I >>think Bob has done a great job with Crafty! :) I also think >>that with the popularity and ease to get source/opening book/ >>learning.dat information, most programers use Crafty as a >>testbed anyway. This means that they spare with Crafty all >>the time and improve as a result (Crafty gets a lot of play >>with the others on the servers, but not with the latest/well >>prepared versions). > >Crafty may play against the other programs on the servers but these are >not programs operated by anybody connected with his competitors. They >are just people who have bought the programs. > >I don't accept that there is heavy tuning or training games played >against Crafty. Most training will be against the usual commercially >available programs and done via automated interface. Genius, Mchess, >Hiarcs, Rebel, Fritz etc. Not Crafty. For example, I've never played one >game here against Crafty, Thorsten has domn a few, but these were for >himelf, and the games and/or conclusions never got passed to me. > >Its just not true that Crafty gets used as a testbad; maybe by some >relatively unknown amateur programs, but certainly not by the major >programs - their opposition is not Crafty. > >If you need to try and generate an excuse, anti-Crafty tuning is not >going to be it. > I have no idea about this and it really doesn't matter. However, on the other side of the coin, I doubt there is a *single* commercial programmer that hasn't looked carefully at crafty. This can be discerned from comments posted here and there where people point to a particular idea or whatever and reference the source code. Which is perfectly ok by me... I'd not try to generate "excuses". It didn't play well. I am becoming convinced that a lot of my eval tuning on ICC hurt it at the very long time controls (compared to ICC) in Paris, because it is *very* sensitive to being attacked, and in the two losses I looked at it was trying its best to avoid getting attacked when it really was in no danger because of the way programs "try" to attack (for those few that even try.) This I'm going to study for a while, and may sacrifice blitz performance against the anti-computer attacker players in order to improve the way it plays in longer games. Whatever happened, however, was clearly *my* fault. not the fault of anyone else. > >> Not sure how Bob can fight back? Something >>we should all think about this year and perhaps help out with >>(perhaps a three month blanket period for Bob/Crafty to prepare?) > >Bob needs new concepts and ideas, mainly in the search. His evaluation >function is probably ok, given that he wants to stay as a fast program, >massive evaluation additions are not feasible. Its the search. His ideas >are behind the other fast programs. Simply getting speed by 64 bit stuff >obviously isn't enough, its algorithmic search improvements that he >needs. > >Chris Whittington I haven't seen a "search" problem at all, against any opposition. I see some serious eval problems in several places, one being the overly sensitive king-safety code. But I generally don't see it drop material, nor fall to cheap tactical tricks (other than to the occasional trapped piece or pinned piece which is really not a search issue so much). however, it can play chess. And it will get better. Time will tell just how much better. But I believe it was playing better (against computers) last year in Jakarta than this year, because so much work has gone into the anti-computer problems that are so apparent on ICC. I don't believe that any of the programs in Paris can cope with some of the players on ICC any better than Crafty already does, because of this king-safety issue. But, of course, that is not going to help much against computers, which don't attack anyway. It now becomes a matter of what to do here. Improve its play against computers, or continuing to improve its play against humans, who are becoming more and more aware of the anti-computer strategies and long-range attacks that programs understand hardly at all. If the WMCCC was my only target, I'd do things a lot differently. If Human games was my only target, I'd stay on my present course. It's an interesting question. I'd love to see a couple of the commercial programs do an automatic interface and play on ICC for a couple of months to establish a rating. Then we could have a multi-round-robin match between the programs to see how they do against each other, and compare that to their human-ratings on the server. I think that would be quite interesting, as it could either support or disprove my thinking that playing against computers and against computer-aware-humans are not the same problem at all. If you want to see how you handle slow-building attacks, try "greg1". CSTal might do very well. It might get rolled into a wad. I'd like to see which, because this is a serious issue for programs that are exposed to humans in an "open" way like on the servers...
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.