Author: Robert Hyatt
Date: 10:33:11 11/13/97
Go up one level in this thread
On November 13, 1997 at 10:06:34, Chris Whittington wrote: > > >Yeah, well, beligerent Hyattian response expected :) > remember, it was not *I* that came up with the "crowing" explanation... I'm just pointing out that just because you finished a few places above me *this* year doesn't mean you are better or worse. There are far better ways to answer that question. I had my own problems in Paris, caused solely by my programming decisions. I simply take my lumps and plan on doing better next year. I'd certainly be willing to put together a small internet-based tournament if you'd like, maybe with 8 programs, 4 fast and 4 slow, and see how things turn out. For Slow, I'd suggest (maybe) CSTal, Hiarcs (I think it seems to be slow based on NPS), and I'm not sure who else. For fast, I'd suggest at least Ferret, Crafty and Fritz would fit pretty well. Speeds I have seen suggest Junior is fast, at least it was the last time Shay was running it on ICC, when comparing NPS figures. Rebel is certainly fast. I'm not sure where to put Genius, but it seems fast. Want to try a "fast vs slow" team event, maybe where each of the 4 fast programs plays a double-round-robin against each of the slow programs? It would at least provide some insight into the fast vs slow issue. If you are interested, maybe we can start a discussion here for the 4 programs on each side? Nothing says Crafty has to be in this, because it might not be one of the best 4 fast programs we can put together... Anyone interested?? >Have a beligerent response back - knowledge based programs are going to >win out in the end. You've only just seen the start of it. The Hyattian >quiescence paradigm is outdated, unoriginal, devoid of new ideas, and >has had its day. > I think you are just a tad optimistic. I'd rephrase that last clause to read: "and has had its day in the past, and will likely have its day again in the future." If you don't believe that, all I can say is "look out." You are playing in heavy traffic. There are some *big* trucks running back and forth. All it takes is one hit... you want some analysis, here is my perspective: I am on the fast end of things. But that does *not* mean that Crafty won't become smarter over time. More on this a bit later however... You are on the slow end of things. Slow != Good. Fast != Good. I am on the side of what I will call "positional chess"... and I read "My System" 30 years ago and haven't forgotten it yet... pawn structure is most important in winning the majority of the games played. You might study a lot of Tal's games, and find many where he attacked and won, but you also find many where he won nice endgames. You are on the other side of this "hill", striving to follow paths that lead to complex and unclear positions. Nothing wrong with that. But when you do it while wrecking your position, it can and does backfire. To play with and beat GM players, you are going to have to retreat quite a ways from where you are. Your program plays too wildly, which will win some games. but it is going to leave itself positionally lost in all the others. And that is not good enough, IMHO. I have a ways to go too. You don't like my current search. That's your opinion. I haven't gotten outsearched in quite a while, haven't made what I would call a tactical mistake in quite a while, when considering the current speed and depth I get. So I'm reasonably happy with where I am. I intend to play with singular extensions again, later, which may (or may not) help with some deep tactical lines, but I can do that without throwing out what I have already done... So I'm climbing the hill up the front, improving my pawn structure eval bit by bit, improving my understanding of piece coordination bit by bit, and I'm going to continue to get better, indefinitely. You ran over the top of the hill, and so far down the back side, that you have problems with programs that your attacking style can't crush with the complications. And you've slowly started coming back toward the top, getting rid of some of the unsound sacs. A disinterested third party might notice that we are both moving toward the *exact* same goal, but from opposite sides. Who gets there first is not easy to say. But there's nothing *at present* that would convince me that you arrive before I do. You may well do so. And you might not. Based on Thorsten's remarks about playing the two programs, my approach seems to have its plusses. About fast vs slow. What's fast? 80K is my speed on a P6. That is also Rebel's speed. We are both *way* behind Fritz. You are at 4K. You are definitely slow. But that does *not* necessarily mean you are better. I recall a program named Awit, by Tony Marsland, which finished in 2nd at the 1983 WCCC in New York, searching 100 nodes per second. I was doing 20K and won, Ken Thompson was doing 160K and came in behind me and Tony. But ask Tony and he'd hardly say that he was better than Belle. There were really 3 contenders that year, Belle, Cray Blitz, and Nuchess. The others were a ways back. So, again, slow is != good, and more than fast == bad. It isn't so much how fast you go, but more of what you do while going that fast. I'm getting slower every year, because of the eval which is now right at 50% of the total search time. This might hit 60-70% by next year. I don't count. I only work on doing what I think are the "right" things (knowledge-wise) and try to do them as quickly as I possibly can. >Viva chess knowledge. Bits, bytes, 64-bits, knowledge independant >null-move bollocks - go take a running jump :) we will take that jump. And I bet you feel *exactly* where we land. :) After you pick yourself up and dust yourself off, that is. :) > >Chris Whittington Back to the knowledge question: You are worried about following a pathway into unclear territory. I want to do the opposite... I want to find a pathway into winning territory. A comment I have gotten quite frequently for a year or so now concerns the "outside passed pawn" evaluation I do. Players have begun to notice that Crafty (a) likes to reach positions where it has one, even if it is (on occasion) a pawn down; (b) it then seems to understand that trading down makes that pawn more important; and (c) it also understand that it has to give it up at the right moment. none of that is complicated. But it wins about 1/3 of its games based on that code, which is fairly complicated overall. There are many other such ideas that I do, some that I don't do (yet). For example, I don't understand that a queen-side majority is really an outside passed pawn, *yet*. I'm working on that now. But I've not seen many "fast and dumb" programs that make more positional material sacrifices than I do, maybe *too* many in fact. And I'm working to stop most of that without taking away the important aspects of the knowledge that is causing them. I don't see *anything* that limits my current approach. I don't see *any* large barrier that lies ahead and blocks progress beyond a certain point. So I think you write me (and others) off too quickly. You might do well with your approach. I *know* I'm going to do well with mine. The current best program in the world is nothing more than fast + knowledge. I can do both, but I can't do it as fast as they can. *yet*... But "never" is a very long time... :)
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.