Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Hsu Presents a Paper at

Author: Don Dailey

Date: 10:42:08 06/26/98

Go up one level in this thread


Hi Bob,

I disagree on your disagree :


>>One  more  example.   Endgame   databases are an   example  of PERFECT
>>knowledge  and  most databases contain  thousands   if not millions of
>>terms or  parameters or whatever  you  choose to  call them.   And yet
>>despite this massive amount of knowledge,  they do not contribute more
>>than a few rating points to a chess program.  And this is with PERFECT
>>knowledge.   Most of the strength   they do contribute is concentrated
>>among 2   or 3 common  endings.  It's  a  whole lot of  knowledge, but
>>rarely used.
>>
>
>I disagree here.  KRP vs KR is a big winner.

I'm not sure what you are disagreeing about.  This is one of the
endings that I believe is in the 2 or 3 common ones that should be
covered.

> I win many pawns, against
>GM's and more particularly against other programs, and many of these games
>end up KRP vs KR where I win when ahead, and draw when behind.  A couple
>of weeks ago, Crafty was playing DiepX, and in 4 successive games, 3 were
>won by this endgame database.  Because I won a pawn, held on to it, trading
>into a won position at just the right moment.  3 of 4 there.  There is hardly
>a day (playing other computers) where I don't see several of these.  Against
>humans, no, because they break badly when they break and the game doesn't
>drag on...

I actually had a suspicion  you would write back saying you won a large
percentage of games as a direct result of some database.  But instead you
said you won a LOT of games with them.  But I need to know what percentage
of games you won.  But before you tell me this, I also need to know
how many of these games you would not have won without the database.
This ending can be won by most programs most of the time without the
database.   I know this because I once asked Larry to help me cover
this ending with knowledge and we discovered it was not difficult to
do.  I'm not saying this is just as good, the database is clearly a
better way to handle this and has the advantage of knowing the
game theoretic results in advance.  But what I am saying is that
you do not really know how many games it's actually winning for you.
I don't think you can claim to be winning games left and right because
of this database.

I don't discount the value of databases at all, I'm for them and we
have a student working on them for Cilkchess.  I have to be realistic
though and I don't expect anything more than a few rating points, I
would say even 30 points is wildly optimistic.   Think about what
30 rating points really mean.  You have to pick up a half point
pretty often to get even 30 rating points.  A lot of games never get to
endings, the ones that do usually get to ones that are draws, or
easy wins where no database is needed.  Now a small percentage of what
is left over end up in one of our databases but most of these databases
are easy wins (depending on which ones you implement of course) and
the program again doesn't really desparately need them (I don't need rook
vs king.)
The ending BNvsK is hard to win without specific knowledge or a
database but it occurs so rarely that I doubt it's worth more
than a single fraction of a percent in rating.
Finally you get to the interesting endings that occur frequently
compared to other endings (but still not very frequently) where
you must win or draw and it's not completely trivial.  Still
most programs will manage to win or draw them.  Your example is
a case in point,  a little knowledge can win this ending (most
of the time) without needing the database.

Bob,

All of my discussion above assumes the current state of the art in
endgame database technology.  If it was possible to cover every
single 6 and 7 man database, then I think we would have a real
breakthrough.  Programs would suddenly be 40-80 rating points
stronger and opponents would be pressured to avoid simple
endgames and indirectly more complex ones.

Long live "brute force"!!



>>There will quickly come  a point when  the bad knowledge that ALL  our
>>programs have will  place a bound  on its  possible strength (given  a
>>certain amount of hardware.)  Learning lots of rare endings and how to
>>play them will help in tiny incremental ways but never make up for the
>>other problems  that  are   more  QUALITATIVE.
>>
>>Saying Deep Blue or any other program has a billion parameters in it's
>>evaluation function  doesn't say much of anything  about how  good its
>>evaluation is.
>>
>>Bob makes a strong  case that Deep  Blue is not limited to  simplistic
>>evaluation just because it is implemented  in hardware and in fact may
>>contain more terms than any other program.  This  is great and I'm all
>>for it, wish  I had it too.  But  it's really hard for  me to draw the
>>conclusion that it MUST be  better because of  this.  This is an issue
>>that must be considered separately  I consider the  term count a minor
>>point although not totally irrelevant. (Don't blast me Bob because I'm
>>not saying Deep Blue sucks either,  I  certainly appreciate  the point
>>you  are  making with Vincent that hardware evaluation  does not equal
>>STUPID program.)
>
>
><blaster holstered>  :)
>
>I have no idea of a way to prove that a bigger eval is better, but I'd
>certainly bet that it is no worse, unless it's written by dummies of
>course...

My point exactly.  I don't have a problem with more and more knowledge,
I'm just saying the real problem is in the engineering of this
knowledge.  I think the bottleneck is this.


- Don



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.