Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Analysing while retracting moves

Author: Robert Hyatt

Date: 18:31:04 11/25/01

Go up one level in this thread


On November 24, 2001 at 22:02:56, Tom Kerrigan wrote:

>On November 24, 2001 at 20:43:22, Robert Hyatt wrote:
>
>>I suppose you missed the point of the discussion?  Searching from back to
>
>Hmm.
>
>>I'm not saying _either_ is better or worse than the other.  I'm simply saying
>>front-to-back is _consistent_ while back-to-front is not.
>
>Funny, you said a lot about accuracy in your original post but nothing about
>consistency. Maybe I missed the point of the discussion when you decided to
>change it without telling anyone?

NO... just read it again, and _carefully_ this time.  I have been completely
consistent.  front-to-back won't see the "error" until it is within the
event-horizon of the search.  This reports the error at point "A" every timee
you try it.  Back-to-front moves the error report back from A, closer to the
original error B, but unfortunately it will be at some random point between
A and B.  And _that_ is the thing I didn't like when I did it this way.

Change the hash table size, the error report shifts around somewhere between A
and B.  Change the search time limit per move, ditto.  Depending on critical
info to remain in the hash table is not the way to try to debug a program, nor
is it interesting once you deliver a program to tens of thousands of people
and have them ask "why is this going on???"

Happened to me once...



>
>Re-reading your original post, you seem to believe that neither approach is able
>to find mistakes, which I think is already supremely wrong. Then you go on to
>imply that one wronly indicated "mistake" is somehow better than a different,
>wrongly indicated "mistake," which, I suppose, is where your recently proposed
>consistency argument comes into play.

No, I didn't say neither approach is capable of finding "mistakes".  You are
again simply missing the point.  We are talking about mistakes that the program
can _not_ find on its own, starting a search at the point where the original
real mistake was made.  As a result, the program will "report" the mistake
somewhere down the line when it finally reaches a position close enough to the
mistake to recognize that something is wrong.

_that_ is the kind of mistake we are talking about catching.  You make a bad
blunder at move B in the game.  When the program searches that position it
can't tell it is a blunder.  It won't see that it was a blunder until the
score plummets at move A further on.  And it reports the error there but it is
wrong.

If you go back to front, the error report gets backed up randomly to somewhere
between the original move B and the move it saw the score drop at A.





>
>-Tom


The question is do you want the program to consistently find the error at the
same point each time, or do you want it to _sometimes_ find it a bit sooner?
Even though _both_ errors it "finds and reports" are meaningless as the _real_
mistake occurred many move prior to either error report...



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.