Author: Peter McKenzie
Date: 21:58:25 08/15/02
Go up one level in this thread
On August 16, 2002 at 00:00:32, Russell Reagan wrote: >On August 15, 2002 at 22:42:26, Peter McKenzie wrote: > >>I pretty much disagree with everything you said. >> >>Its perfectly possible to be super strong without any big secrets. Attention to >>detail, improving the little things, and a whole lot of blood sweat and tears >>will get you a very long way. Oh, and did I mention testing? Good testing >>helps heaps. >> >>5% inspiration, 95% perspiration :-) > >I like, "No pain...no pain!" :) > >>Of course, some commercial engines have big secrets, but I don't think they are >>necessary to be super strong. > >What I said shouldn't apply 100% to 100% of the all commercial engines. > >A perfect example is that Junior does not use null-move forward pruning. There >is no way possible to take well known techniques, subtract null-move, and be one >of the best in the world. Conclusion: "big secrets". I'm not sure Junior doesn't use null move, not that it matters much. My vague idea is that they have a very flexible way of counting plies. So captures and various active moves increase the ply count more than stupid looking moves. This isn't a particularly secret idea, although the details of their implementation are probably quite secret. I've also heard rumours that Junior uses a cut-down q-search, or maybe even none at all. Again, this isn't a new idea but if implemented right I'm sure it could be effective. It could be synergistic with their basic approach to search. > >Tiger doesn't use null-move to a large degree, and makes use of other forward >pruning techniques. I imagine Tiger does normal null-move pruning except near the tips where it does something more aggressive. I agree that exactly what he is doing near the tips is indeed a "big secret". It seems likely to me that he is doing some sort of static based (as opposed to search based) pruning based on chess specific knowledge. > >IIRC, Rebel doesn't (or didn't) use null-move in the past, and was still among >the strong commercial engines. Yup, but Rebel uses null-move now, probably uses his own pruning near tips like Tiger does (or so I think). > >I don't know about Fritz, but IMO if you take things that are well known and put Fritz uses nullmove pruning. Maybe it uses other pruning too. I think Fritz gets much of its strength from being extemely fast and well tuned. >in a ton of work, you might surpass Crafty, but I think you're still only going >to be at a level of Gandalf, which seems to be a step or two behind the Fritz's, >Junior's, etc. of the computer chess world. We agree to differ then. I believe that with known searching techniques and a *really good* evaluation you can make something that matches commercial programs. Note that crafty only uses 1 particular combination of known search techniques, there are things that Bob doesn't like or hasn't invested much time in. > >I would find it very hard to believe that there aren't quite a few significant >secrets in the commercial programs. Sure, it's possible that *some* don't have >any big secrets and simply make use of well known methods which have been >refined for many years, but even if there are a few of those, there are three I >could think of off the top of my head that must make use of some "big secret" or >else they would not be as good as they are. I mean if you take the well known >methods, subtract null-move, you're looking at maybe a branching factor of 4-5 >at best, **UNLESS** you are using some other method that is not public >knowledge, AKA a secret. The top engines consistently have branching factors >between 2-3, and since we know that some either use null-move to only a small >degree or do not use it at all, we know that there must be something that they >are doing to achieve such great results. > >I'd like to know what you think about all of this. If you are looking for big secrets which can be easily added to any program with a gain of 50 rating points, I fear there aren't many of these. I think a program's strength comes from a whole combination of things working well together. cheers, Peter > >Russell
This page took 0.01 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.