Author: Robert Hyatt
Date: 13:11:56 12/20/03
Go up one level in this thread
On December 20, 2003 at 05:03:04, martin fierz wrote:
>On December 19, 2003 at 14:41:11, Robert Hyatt wrote:
>
>>On December 19, 2003 at 14:33:08, martin fierz wrote:
>>
>>>On December 18, 2003 at 09:28:13, Robert Hyatt wrote:
>>>
>>>>On December 17, 2003 at 18:20:09, martin fierz wrote:
>>>>
>>>>>On December 17, 2003 at 12:38:45, Daniel Clausen wrote:
>>>>>
>>>>>>On December 17, 2003 at 09:35:16, Robert Hyatt wrote:
>>>>>>
>>>>>>[snip]
>>>>>>
>>>>>>>And, as I suggested previously, if, after a program leaves book, it is
>>>>>>>in an obviously won or lost position, the game gets aborted and the next
>>>>>>>one started. There is no place for "book kills" when the goal is a time
>>>>>>>handicap match.
>>>>>>
>>>>>>In order to reduce threads like 'this opening position is lost! no it's not! yes
>>>>>>it is! it's lost when you use bitboards! but fisher would win this position vs
>>>>>>DB!' it would be good to 'formalize' won/lost positions after the opening.
>>>>>>
>>>>>>You could declare an opening won/lost if one of the engines evaluates its first
>>>>>>move out of book with a score outside a predefined score-window [X, Y]. ([-0.5,
>>>>>>+0.5] could be an example) Some points:
>>>>>>
>>>>>>- I intentionally used two variables so it's possible to have an assymetric
>>>>>>window (no clue whether that could be helpful or not)
>>>>>>- You don't capture positions where a human being with comp-chess knowledge
>>>>>>knows, that one engine _will_ lose but the scores of the engines won't catch it
>>>>>>- the scores for this score-window have to be adjusted (+1 should mean approx 1
>>>>>>pawn advantage)
>>>>>>- everything else I forgot :)
>>>>>>
>>>>>>Ideas? Comments? Shrieking epitaphs?
>>>>>>
>>>>>>Sargon
>>>>>>
>>>>>>PS. It's funny - we often claim that "{small number} games are not enough!" but
>>>>>>now we still make this experiment ;)
>>>>>
>>>>>
>>>>>why not just run a nunn / nunn2 match, or select a few openings for this match
>>>>>to be played with both sides? this would remove the book dependence of the
>>>>>match.
>>>>>
>>>>>cheers
>>>>> martin
>>>>
>>>>
>>>>well-known positions are bad. Too easy to tune specifically for them, which
>>>>is a problem.
>>>
>>>let me repeat, this time with CAPITALS:
>>>
>>>why not just run a nunn / nunn2 match, OR SELECT A FEW OPENINGS FOR THIS MATCH
>>>to be played with both sides?
>>
>>I can play when I know the rules. :) So I will respond, highlighting the
>>important part:
>>
>>"Why not just run a nunn / nunn2 match"
>> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>>
>>
>>I responded to that part.
>
>of course - i could see that. only when you do that, it gives the impression
>that you didn't read my original post. you could say: "playing nunn matches is
>bad, because they may have been tuned for, but selecting some positions for this
>match is ok".
>then i would have the impression you read my post and we wouldn't have to resort
>to CAPITALS or underscoring :-)
> ^^^^^^^^^^^^
>cheers
> martin
OK. I'll take credit for not being very clear. What I was saying was
"don't use the nunn-type tests." I could have said "don't use the nunn-type
test, the random position idea is ok."
Bob
>
>"Well known positions are bad." That was all
>>I responded to. "select a few openings" would _obviously_ not be "well known
>>positions..."
>>
>>
>>
>>
>>
>>>i understand you could tune for a nunn/nunn2 match, although i doubt that any
>>>such tuning would be very effective. tuning for a single test position, very
>>>easy. tuning for a test set, harder. tuning for a multitude of game positions
>>>where there is no solution, very hard IMO. if your program really plays better
>>>in the nunn2 match thanks to some tuning in 2x20 games, then i guess it is
>>>simply a stronger engine.
>>>
>>>if you get somebody to select a few openings, then nobody can tune in advance.
>>>
>>
>>that last sentence was all I was suggesting. The first part of your
>>suggestion had what I considered to be an important "loophole". There have
>>been _plenty_ of cases of programmers tuning for particular positions,
>>or particular openings, or whatever...
>>
>>
>>
>>
>>>cheers
>>> martin
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.