Author: Robert Hyatt
Date: 13:19:33 04/19/02
Go up one level in this thread
On April 19, 2002 at 12:54:47, Gian-Carlo Pascutto wrote: >On April 19, 2002 at 12:31:46, Robert Hyatt wrote: > >>This is wrong thinking. If it will play the right move for the wrong reason, >>then in similar positions it will play the _wrong_ move for the wrong reason, >>and lose. >> >>All you have to do is add a simple endgame rule that says _always_ centralize >>the king to see why this happens. Then give your opponent a passed pawn on >>the edge of the board and see how much that king in the center helps. The >>idea is that centralizing is not the right idea. The right idea is to get the >>king to wherevere it is needed, whether that is the edge, the corner, or the >>center. Just moving it to the center will be right in plenty of positions. >>But it will be wrong in enough to make it obvious, too... > >Good example. Having the king in the center in the endgame is an excellent >rule. It will be right much more than it will be wrong. You'll easily see >a significant different in playing strength when comparing programs with or >without this rule. OK... I will play your program and all it can use is "centralize the king". Care to bet on how endgames are going to turn out? Hint: I _won't_ be making passers in the center. That is a weakness that can easily be exploited once it is spotted. It might work fine in most _random_ positions. But I am not going to give your program random positions. I'm going to give it positions where I know it will screw up. And it will. Over and over and over. That is why it won't work... I used to do this early on. And I was "found out" and had to fix it so that the thing knows where the king needs to be and gets it there quickly. > >Sure, it won't always be right. But it'll be right a hell of a lot of times >and do much better than one that doesn't have this rule. Not when it plays _me_. Or another player that I talk to. :) > >If you play a program with and one without this rule and the one with the >rule wins the endgame time and time again, do you still call that luck? After >all, the rule could be wrong, so hey, the program is just lucky it was right >all those times, because it doesn't really understand what's going on, is >it? > >I don't think this kind of reasoning 'holds'... It holds just fine in the real world. Please play Crafty using such a rule, because crafty specifically has code to try to produce passers on either wing rather than in the center, for obvious reasons. And your "centralize" rule will fall flat most of the time. Granted, if your opponent doesn't know about passed pawns (distant passed pawns) and if he doesn't know you have this incredible weakness about where to put the king, you might do quite well. But against some programs, this will fail. Against _any_ "good human" it will be devastating... > >>> You can gamble, take risks, and make sure >>>that you gamble better than your opponent. >>> >> >>If you take this approach, then beating the top GM players will be totally >>impossible. You can't rely on luck. > >Not at all. If your program estimates the risk well enough and gambles at >the right times. It will win. You're saying a program as Junior can't compete >with top GM's because of this? I didn't say that at all. We were talking about _this_ position. And I clearly said that if a program makes the right move for the wrong reason it is _not_ a good thing. Even though it turns out right in "random" positions. But once that weakness is spotted, humans will give it the "wrong reason" every chance they get... > >>>Chris Wittington realized this first, and Christophe Theron was next. Junior 7 >>>also follows this strategy. If you take a closer look, you'll see that all >>>top SSDF programs have elements of this playing style. >> >><sigh> Another "new paradigm" thread? I don't think so. Comparing CSTal >>with any other program is pointless. They are not the same. > >I specifically wanted to avoid that wording. But you can't deny there's >been a shift towards more speculative attacking play, and it's been pretty >effective. This has been going on for 20 years. Just drag out your old Novag Super- Constellation and watch it make speculative sacrifices that are reasonably sound most of the time (against typical players). > >>>It's a design decision to make Junior play moves like Nxh6 without fully >>>understanding or calculating where it's going to end up. Sometimes it will >>>backfire. If you look at Juniors performances in tournaments, you'll realize >>>it works much more often than it backfires. >>> >>>Junior gambles because it knows the odds are in it's favor. If you win >>>a bet where you had, say, >95% winning chance, were you really 'lucky'? >>> >> >>if it was _really_ 95%, no. But is it _really_ right that often? I doubt >>it. > >If it's > 50%, that's good enough for me, because it will start winning >games over programs that don't have it. But what If I force it into those 49% of the positions where it screws up. And what if I do that 100% of the time when it plays _me_. I don't know if you have spent much time talking to GM players, but believe me, once you do, you will appreciate just how much they "spot". This has certainly helped me many times to make small improvements that produce good results... IE 30 years ago every blew programs away because programs didn't quite understand passed pawns. They would let a human trade down to an ending where the program was multiple pawns ahead, but the human had one that could not be stopped but the depth wasn't sufficient to actually see the promotion. And humans did this over and over and over... and they won and won and won. If you have a weakness, and it is spotted, it will become an achilles heel until it is fixed. ICC is quite brutal in this regard... > >-- >GCP
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.