Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Nolot Positions #1

Author: Robert Hyatt

Date: 07:30:00 11/28/00

Go up one level in this thread


On November 28, 2000 at 02:35:39, Bruce Moreland wrote:

>On November 28, 2000 at 00:12:09, Robert Hyatt wrote:
>
>>On November 27, 2000 at 22:10:07, Mogens Larsen wrote:
>>
>>>On November 27, 2000 at 21:52:40, Robert Hyatt wrote:
>>>
>>>>What I was referring to was playing the right move with a score that is < 0,
>>>>when the move is a _winning_ move.  If the program _knows_ it is winning, it
>>>>will have a positive score...  If it doesn't know this, it is playing it for
>>>>the wrong reason.
>>>
>>>That's just a question of depth since the score is obviously improving. When a
>>>program is improving its score by chosing the right moves (to a certain extent
>>>as inaccuracy tends to creep in the further you go in the PV), then I consider
>>>it to be the right reason.
>>>
>>>If Nxh6 had been the only correct move in the PV, but the score > 0, then it's a
>>>right move for the wrong reason to me. The score will improve if the PV keeps on
>>>approaching the correct variation. You're doing a TC imitation by staring at the
>>>score IMHO.
>>>
>>>What score would imply "right reason" at depth 11?
>>>
>>>Mogens.
>>
>>
>>I disagree.  There have been multiple cases of programs being specifically
>>tuned to play certain moves in test positions.  The PV was "close" but missed
>>the important move that made the score _really_ improve.
>>
>>Whether that happened in this case or not, I have no idea.  But when someone
>>says "white to play and _win_" and the program plays the right move but with
>>a score that says "if I could find a repetition I would take it in a heartbeat"
>>then I remain skeptical of its ability to really understand what is going on.
>>And suspect it is very likely that it may well go wrong...
>
>Why do we have some weird controversy any time this stuff gets discussed?
>
>1) There has never been any evidence presented that anyone has tuned for any of
>these Nolot positions ever.  Other test suites perhaps, but Nolot, no.

I didn't say anybody had.  I said that this issue has come up often enough to
make me skeptical of results that have the right move, wrong score.  I have the
same criticism for my program too, of course.  But white to play and win does
not mean a score of -.2 shows it knows what it is doing...



>
>2) Many of these positions have been found by computers for a long time,
>including this one.  This is one of the easier ones to "see".

To use a Clintonism, "it depends on what you mean by found."  There are too
many examples of a program finding the right move, with the wrong eval, then
changing to something else after a longer search...  If the program sees +4,
then it likely won't change unless it finds something even better that is also
going to win convincingly.  But if it says -.2, it is likely that in a tactical
thread, it will veer off to take a perpetual, since draw would be better than
-.2...




>
>3) I have never seen any "winning" PV in this position produced by a program.
>Everyone finds this eventually, with a minus score.  Hsu says they found a win
>with DT, but doesn't provide the score.

With no winning score, I wouldn't say it is "solved".  It might not be winning,
or the program(s) might not be able to find the win.  Just finding the right
move is a dangerous thing...   if it isn't for the right reason.



>
>4) I don't think that anyone made any claims that any program was Joe Super-Stud
>because it found a "draw" or "surviving" PV in this position.


I didn't make any claim either, other than that it isn't very convincing
that a program plays the right move but has a bad score...




>
>Within the past month or so, you have insinuated dishonesty or evil intent a
>number of times, or at least you've brought up the issue and directed several
>threads away from their initial topic and on to issues of the integrity of the
>individual in question.  You've done this with little or no evidence beyond the
>considerable weight that your name carries.  And when it has been pointed out to
>you that you've done this, it has been extremely difficult to get you to
>retreat, even the slightest bit, from these damaging insinuations.  And as far
>as I can tell you've been wrong in every case.


I don't know what you are talking about in this case.  I didn't imply any
dishonesty, nor accuse anybody of _anything_.  I simply said that there have
been so many cases of right move wrong score, some due to hand tuning for a
specific problem by a specific author, some due to simply misunderstanding the
position enough to make the right move for the wrong reason.

As far as the "damaging insinuations" I only recall one thread where this
came up.  Can you point to others?





>
>There is no logical reason for you to bring up this topic with regard to this
>program and this test position.  All you are doing is haphazardly smearing the
>Gandalf author, and I have no idea why you could mean to do this.
>
>bruce




I didn't intend to smear _anybody_.  I did intend to point out that if a program
chooses the right move for the wrong reason, I don't consider _that_ solving
the position.  Nothing more.  Nothing less.  I don't see your point here, and
I don't see how anything I said would "smear the author."




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.