Author: Robert Hyatt
Date: 08:56:38 11/16/98
Go up one level in this thread
On November 15, 1998 at 07:09:47, Amir Ban wrote: >On November 14, 1998 at 19:07:55, Thorsten Czub wrote: > >>On November 14, 1998 at 14:04:46, Ed Schröder wrote: >> >>>>Others (who have also bought junior5) have found out the same. >> >>fact, these persons have posted here. >> >>>>Ed claimed the same very early. >> >>fact too. ed said that the fritz5 autoplayer stops rebel from learning, posted >>in old discussions months ago. >>i have not said that you claimed something about the junior autoplayer ed. >>i said you claimed the same, meaning you said that rebels learning was not >>working in your observation months before. but anybody oversaw this. >> >> >>>I claimed nothing. Junior5 even hasn't been installed on my autoplayers >>>so I can't know. >> >>i have not said you claimed about junior. junior is only an engine in the fritz5 >>user interface. what happens to junior might have also happened to the secret >>chessbase fritz5 autoplayer. and you proofed this autoplayer, or ? >>you had a log-file of the data send from ONE to Other machine, or ? >>And there was nothing in the data. all clean. only that rebel was unable to >>learn. or ? I remember that this was said months ago. >> >>>- Ed - > > >The F5 autoplayer and the J5 autoplayer are not the same. Porting the Win16 >autoplayer to Win32 was too much work, so the J5 autoplayer was rewritten, >based, I believe, on Stefan MK's code. > >It's not yet clear what this J5 autoplayer problem is, and whether we are >dealing with a single problem or several unrelated ones, but I don't understand >what this has with saving the opponent's game. In the past, ChessBase >acknowledged a problem in the F5 autoplayer's saving of opponent's game, but at >least apparently this is working ok in the J5 autoplayer. > >If a J5 autoplaying session hangs or crashes in some conditions that are yet >unclear, to conclude that this happens deliberately to handicap the opponent is >a weird and hostile interpretation. > >Amir I would take this one step further. I can't see how, *even if* Junior doesn't do something exactly right, how one could blame Junior (or Fritz in this context) for what I would argue is a bug *in the other end*... IE why would *I* write a program that depends on my opponent doing something before *my program* learns? That's a poor design to start with. Because then we get into such arguments as what happened back when Fritz jumped to #1 on the SSDF list. I'd worry about *two* issues here: (1) my opponent might try to "fudge" and not tell me the game is over so I can "learn" and (2) my opponent might have a bug so he wouldn't know he should tell me this. My solution has always been to fix things on *my* end so I don't have to depend on anybody to tell me what to do. I ran into this early on ICC... where folks found out they could disconnect and then restart and if they did this within 10 moves of the last time, crafty would *never* learn as I initially waited for 10 non-book moves to figure out what was going on. And the human players figured this out too and would find a book bust, and disconnect every 8 moves or so to prevent crafty from learning that it was bad. :) It was fixed shortly after that. And I believe that is the way to write this kind of software so that your opponent can do *nothing* to break your code. You (not Amir here) can either grumble about your opponent not exactly following a protocol or you can take action so that no matter what he does you do what you want/need to do. Then none of this would happen... But it seems to me that it is up to each programmer to cover his own butt in this manner, rather than insisting that everyone *else* must follow the rules. In that way the program works as expected, all the time, regardless of a buggy or intentionally deceitful opponent misbehaving... Ie isn't this the purpose for book learning in the first place? To avoid playing a losing line repeatedly? Then make it work 100% of the time as it should...
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.