Author: Bruce Moreland
Date: 09:39:13 06/29/98
Go up one level in this thread
On June 29, 1998 at 00:30:49, Robert Hyatt wrote: >On June 28, 1998 at 23:48:43, Bruce Moreland wrote: >>But I don't think that the DB project proved anything at all about search or >>evaluation techniques. I don't think that you can look at the project from our >>point of view as outsiders, and say that any technique is shown to be good or >>bad because of what they did. Anyone who tries to say the project proved >>anything ends up making arguments based upon painfully little real evidence, >>most of the evidence is anecdotal *at best*. >perhaps to you. Unfortunately, I have had the opportunity to sit across >the board from them, with a program of known capabilities (Cray Blitz) and >have seen first-hand what they can do. I don't know of *any* program that >has only lost 2-3 games in 10 years of ACM and WCCC tournaments. That is >a *huge* record. I don't know how good they are, to be sure, but I have a >good idea of what Cray Blitz can do, and they are certainly better than it. I am not challenging the record of DT. I am merely stating that it can't be said that null-move is bad just because DB doesn't do it. When a giant comes onto the battlefield, with an 12 foot section of log as a weapon, and knocks everyone's head off, this shouldn't cause everyone else to drop their swords and pick up sticks. The lesson to the little guy is that you should get big. The giant gets no lesson here, since he won, and he'll be back next time with a bigger log. When you introduced Crafty, you had some problems because you didn't do micros, for instance you had some big memcpy's that slowed you down on a micro, but which didn't make any difference on the Cray. But there were some other changes you made because the changes made you stronger on the micro platform. For instance, you went to null move R=2, and you stopped extending so much. I don't think you would have done this if you didn't have competition that could score points against you. If you were running on something twenty times faster than the rest of us had, you'd be doing stuff that you needed to be doing in order to waste wimps. You could talk about exploiting pruning mistakes in the slower programs, you could extend a lot in order to superset the slower machines while still yielding the same search depth. And oh yeah, as I have stated in an old post, you'd be winning all of these games, so by definition everything you did would be working right. You'd feel good about your eval function and your search. Who changes stuff that they feel good about? But if someone came along who could offer you competition, you would have to change, and I bet we'd see null move R=2 again, and we'd see fewer search extensions, and I bet you'd do some muttering about your eval. This is my speculation about DT. I doubt they were placed under much evolutionary pressure, I mean, if you can effectively wield a 12-foot log section, what is going to stop you? DT/DB is a good source of ideas, but if you try to emulate them on a computer that anyone can afford, something really bad will happen. And I think it makes good sense to ask if their project would be a lot different if they'd had someone who put them under some real evolutionary pressure. If you went back to the Cray again, would you really use null move R=1 (or take it out entirely) and tons of expensive extensions? bruce
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.