Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Junior-Crafty hardware user experiment - 17th game

Author: Robert Hyatt

Date: 21:16:03 12/18/03

Go up one level in this thread


On December 18, 2003 at 18:32:21, Peter Berger wrote:

>On December 18, 2003 at 18:18:40, Robert Hyatt wrote:
>
>>On December 18, 2003 at 16:14:58, Peter Berger wrote:
>>
>>>On December 18, 2003 at 15:58:54, Robert Hyatt wrote:
>>>
>>>>On December 18, 2003 at 15:45:17, Peter Berger wrote:
>>>
>>>But I really can't understand how you can say that Crafty likes variety against
>>>computers - with default settings it certainly has the narrowest book against
>>>computers from all engines I have ever seen. And I really think this is a very
>>>bad idea as partially also shown in this match. Thank good that even the 3%
>>>moves sometimes show up eventually - it might save Crafty's head in this match
>>>;).
>>>
>>
>>You missed my point.  I mentioned "ICC".  There variety is _critical_ or
>>the opponents keep finding book lines that lead to trouble for Crafty.  With
>>a wide book, they have a much harder time.  For matches, I would certainly play
>>with "aggressive learning".  As far as "3% moves" I don't know what that might
>>be.  IE I don't (to the best of my knowledge) have any "play 3%" lines in my
>>book.
>
>Did I really miss your point? At ICC some of your opponents are computers too.
>You will face them with the same book as used here - wrong?

Wrong, yes.  First, program A plays crafty a game, and then goes away to
try to cook up something.  Or if he plays crafty (or a crafty clone) and
wins, he moves on to either crafty or another clone and tries that _exact_
same opening again.

Hence my comment about "the book must be wide on ICC".

I've only been playing there for 8+ years now.  I _have_ seen most of the
problems and developed approaches to solve many of them.  Just go there and
play 24 hours a day, 7 days a week, with a narrow book, and see what
happens...


>To call Crafty's
>book wide is really wrong - I will try to reach you the last time with this
>message, aware that I am _most_ probably not the first one to come up with this
>anyway. To answer your second question: given 1. e4 e5, Crafy being white: what
>is the probability that it will play 2. Bc4 against a computer?


10% of the time:

  move     played    %  score    learn     CAP     sortv   P%  P
   Nf3!    152593   81   0.11     0.27  -655.36   14041.0  90  Y
   Bc4!      6913    3   0.07     0.00  -655.36   12046.3  10  Y

that 3% you are quoting is the "percent of the time this was played
in all the games in enormous.pgn, after 1. e4 e5.  Bc4 was played 3%
of the time.  Crafty plays it about once out of every 10 games.

However, once you are past that point, the book is _wide open_.  So
I don't understand your usage of the term "narrow".  IE from here on
in the game the program can choose from _any_ variation in the Bishop's
opening that it wants...




>
>>If you try to use it in an important way, the best idea is to ask.  :)  Then
>>I can tell you how I would do it.  And in a match against a single opponent,
>>I'd go for aggressive learning and put the onus on the opponent to thwart it
>>if it can.  :)
>
>It is not for me to decide if I used Crafty in an important way. I certainly
>used Crafty exactly as you recommended it for this match. If my memory failed
>and I misunderstood one of your recommendations as archived in the CCC files,
>please point out the message I missed - I will simply redo the match.

I don't believe we ever discussed how it should be set up.  I was simply
addressing the "book learning issue".  Nothing more, nothing less.  You can
play it "default" or you can set it for "aggressive learning."  I didn't notice
your asking _me_ specifically about how _I_ would tune it for a match vs a
single opponent.  It was your idea, you chose how to set things up, that's ok
by me.  If I were playing the match, I would do things differently, particularly
with the learning issue.

IE I didn't suggest this match in the first place, so I really didn't give it
any serious thought.  Amir brought up the 100:1 match, which was what _I_ was
originally talking about.  People play matches all the tie.  Ponder=on,
ponder=off, book=on, book=off, crafty book, fritz book, custom book.  I
simply don't pay that much attention to them.  If I am interested in
the outcome, _I_ run the experiment, so that I know what is going on
and why.  Everyone is certainly free to run things however they want,
and to draw whatever conclusions they want.



>
>Peter



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.