Author: Robert Hyatt
Date: 07:32:56 02/21/99
Go up one level in this thread
On February 21, 1999 at 02:20:21, KarinsDad wrote: >On February 21, 1999 at 01:21:50, Robert Hyatt wrote: > >[snip] >> >> >>No... the programs have much _better_ time management than 'humans'. That >>is the point. In a 0 increment game, humans often get into a good position >>but take too long in doing so. With an increment, they can take that advantage >>and turn it into a win. Not real often, mind you, but often enough to be >>noticed... > >Yes, this goes back to my original question to Bruce. > >My effective question was whether it was that the program had an advantage due >to poor human performance vs. blitz (and this is related to time management) or >was there something else. > >In order to clarify, let's take the example of 10 0 vs. 5 5. These times are >very similar in most cases. Therefore, the human should not have much of an >advantage using the 5 5 time limit. And in fact, for a well written program, >it's rating should be just as good (or nearly as good) at 5 5 as it is at 10 0 >(all other things being equal) versus humans. > > You are still missing the point. It is not the _program_ that has the problem here. It is the _human_. And no matter what you an increment increases the human's chances of winning, because even if the screws up in managing his time, he can avoid flagging. in 5 0, 10 0, or 30 0 (witness the ferret games vs blatny) the human will, on occasion, get into great time trouble. Even when he is outright winning. And a certain number of those 'won' games will be lost due to time. In an increment game, won games will be won, usually. Which will have an effect on the rating. >The only place where this may not hold is in the example of a human getting an >advantage and being able to maintain it (due to him having at least 5 seconds >for every move). This is the "something else" that Bruce and you responded with. > >However, I wonder if this is truly the case. if you visit ICC enough you will see this many times... > >To find out, I think one would have to compare 10 0 games from the same human >opponent vs. 5 5 games over a large number of games. Comparing 5 0 and 5 5 >doesn't make sense. Granted, that in a 5 5 game of over 60 moves each, the human >would have more time in which to play and would theoretically have a slight >advantage because of it. But I truly wonder how many times this happens and how >many times the human wins because it happens. Do you have statistics in this >area Robert? I have about 100,000 games that Crafty has played. I'll see what I can dig out of them... > >[snip] >>> >>>Programs can be changed to disallow this as well. They can keep track of who >>>they played and when, and with which opening, etc. The IM played e4 last time, >>>so instead of e6 I will play c6, and next time I will play c5. >>> >>>There are no disadvantages to automated programs, just disadvantages to what is >>>coded within them. >>> >>>KarinsDad >> >> >>your statement is right, but the _idea_ is _wrong_. The entire purpose of >>putting an automatic program on ICC is to play as many games as possible. Not >>to play just a few to protect your rating (at least in my case this is the way >>I think). So I don't try to 'fix' something by avoiding that opponent. I try >>to fix it by making the program respond automatically (such as learning, or >>recognizing the case where my opponent is trying to 'shuffle' and run me out >>of time, and so forth). > >Ok, but having a program play a different (or possibly random) opening each time >(even without recognizing who the opponent is) on a chess server is probably a >good thing. > >KarinsDad This is not so easy. Most opening lines turn out to be bad, if your book comes from PGN games. IE in _any_ opening line you choose to follow, it is almost guaranteed that at 1/2 of the positions, there are good and there are _losing_ book moves to play. If you just choose randomly, you just about guarantee yourself to play one of these in every game. It is very difficult to discover what is good and bad. Which is why 'book learning' is critical. Or else a narrow book which can be predictable.
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.