NIMZO 3 ist out now!
Tom Kerrigan
"ist"? Are you German? :)
>It's a very exiting program, you can modify almost everything.
>The best thing is that you can put in more intelligence with a special
>program language.
Wow! How does that work?? Sounds a lot like what CHESS 4.0 was going to be...
Cheers,
Tom
_______________________________________________________________________________
Tom Kerrigan kerr...@frii.com O-
"Would you tell me, please, which way I ought to go from here?"
"That depends a good deal on where you want to get to," said the Cat
-- Lewis Carrol
Chris Whittington
kerr...@frii.com (Tom Kerrigan) wrote:
>
> "ist"? Are you German? :)
>
> >It's a very exiting program, you can modify almost everything.
> >The best thing is that you can put in more intelligence with a special
> >program language.
>
> Wow! How does that work?? Sounds a lot like what CHESS 4.0 was going to be...
>
> Cheers,
> Tom
>
All the rgcc speed freaks will like it.
It fits the rgcc speed/null-move/quiecence paradigm.
Chess knowledge is pre-processed and fed in to the piece-square
tables.
There's a special pseudo-human language to program the
pre-processing knowledge.
So, if you believe in this way of doing it, you'll like it.
Personally, I think pre-processing has serious flaws. When a few pieces
get changed off during the search, or the position changes in some
way, then inappropriate pre-processing knowledge is still in the
piece-square tables, leading to erroneous evaluations.
The deeper the search the more the difference between ply one
pre-processing knowledge and ply N. John Nunn has criticised Fritz
nicely with an example game from AEGON 95 over this technique.
Chris Whittington
Chris Whittington
hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>
> Chris Whittington (chr...@cpsoft.demon.co.uk) wrote:
> only, since the center of the board can't move around very well. :)
Yup. I think that an important divergence in chess programming style
is shown by this.
1. Pre-process, have very little knowledge processing at each node
and use the piece-square tables.
2. Or don't pre-process, and apply knowledge at each node.
Type 1. programs show very little style range, they will tend to search
the same tree.
Type 2. programs will style range dependant on the knowledge contained
in the evaluation function, I like to think of the evaluation function,
in this case, as a vector which decides which of the (infinite number
of) sub-trees that get searched.
Within type 1. and type 2. all the usual stuff can get used, no reason
why null-move, quiecence (assuming static mtrl evaluator available)
can't be mixed and matched. Type 2. high-knowledge allows more
search ideas to be effectively introduced (forward-pruning, candidate
moves, speculation etc.)
BTW the centre of the board can move around. If the physical centre
gets blocked by pawns then the action-centrer can shift to left or
right.
>
> BTW, you drew the wrong conclusion from my rook scoring. The code to
> attack open files was not to avoid trapping a rook, it was to avoid
> looking foolish. :)
Yup, I know. But it has the *effect* of promoting an 'escape' square
for the rook.
> Example:
>
> Rook at a1, pawn at b2, black attacks the pawn, How to defend, Ra2 or
> Rb1? If the a file is open, Ra2 may be ok... but if not, Ra2 leaves
> the rook buried in a hole. In this case, Rb1 is better because the
> rook still attacks open files by sliding along the back rank. I remember
> one game in particular in a blindfold simul against Mike Valvo years ago
> where we played Ra7 as black, and played the game a rook down as a
> result. I fixed it.
Too be fair, it isn't a 'fix'. Just now the evaluation 'tends' to avoid
such situations. Presumably Kasparov could still force both Crafty's
rooks into dead squares :)
Chris Whittington
> In Crafty, however, the fix was easier with the
> bitmaps, so that there's essentially no cost for doing this current
> "trick." (and not looking like an idiot in the process.) Of course,
> there's still *plenty* of places where Crafty makes me look like an
> idiot, although tactically it doesn't happen often any more as it has
> gotten fast enough to not do gross tactical things... :)
>
> Bob
>
Robert Hyatt
BTW, you drew the wrong conclusion from my rook scoring. The code to
attack open files was not to avoid trapping a rook, it was to avoid
looking foolish. :) Example:
Rook at a1, pawn at b2, black attacks the pawn, How to defend, Ra2 or
Rb1? If the a file is open, Ra2 may be ok... but if not, Ra2 leaves
the rook buried in a hole. In this case, Rb1 is better because the
rook still attacks open files by sliding along the back rank. I remember
one game in particular in a blindfold simul against Mike Valvo years ago
where we played Ra7 as black, and played the game a rook down as a
result. I fixed it. In Crafty, however, the fix was easier with the
Robert Hyatt
Chris Whittington (chr...@cpsoft.demon.co.uk) wrote:
: hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
: >
: Yup. I think that an important divergence in chess programming style
: is shown by this.
:
: 1. Pre-process, have very little knowledge processing at each node
: and use the piece-square tables.
:
: 2. Or don't pre-process, and apply knowledge at each node.
:
: Type 1. programs show very little style range, they will tend to search
: the same tree.
:
: Type 2. programs will style range dependant on the knowledge contained
: in the evaluation function, I like to think of the evaluation function,
: in this case, as a vector which decides which of the (infinite number
: of) sub-trees that get searched.
:
: Within type 1. and type 2. all the usual stuff can get used, no reason
: why null-move, quiecence (assuming static mtrl evaluator available)
: can't be mixed and matched. Type 2. high-knowledge allows more
: search ideas to be effectively introduced (forward-pruning, candidate
: moves, speculation etc.)
:
: BTW the centre of the board can move around. If the physical centre
: gets blocked by pawns then the action-centrer can shift to left or
: right.
:
: >
: > BTW, you drew the wrong conclusion from my rook scoring. The code to
: > attack open files was not to avoid trapping a rook, it was to avoid
: > looking foolish. :)
: Yup, I know. But it has the *effect* of promoting an 'escape' square
: for the rook.
:
:
: > Example:
: >
: > Rook at a1, pawn at b2, black attacks the pawn, How to defend, Ra2 or
: > Rb1? If the a file is open, Ra2 may be ok... but if not, Ra2 leaves
: > the rook buried in a hole. In this case, Rb1 is better because the
: > rook still attacks open files by sliding along the back rank. I remember
: > one game in particular in a blindfold simul against Mike Valvo years ago
: > where we played Ra7 as black, and played the game a rook down as a
: > result. I fixed it.
: Too be fair, it isn't a 'fix'. Just now the evaluation 'tends' to avoid
: such situations. Presumably Kasparov could still force both Crafty's
: rooks into dead squares :)
:
:
: Chris Whittington
:
:
Possibly, but the penalty's big enough that he's going to have to give up
something in return. :) BTW I haven't seen such lousy moves in 10,000
games now, which at least indicates it's "better" if not "best". Before,
it was simply silly. However, there's still plenty of silly stuff to
fix...
Tony Hedlund
>
> All the rgcc speed freaks will like it.
>
It looks very fast. On my P90 it seeks about 34000 pos/sec and 9 ply
deep in the middlegame. I'm a testplayer for the Swedish SSDF and I
don't think I have seen any faster.
Tony
Pitters
In your Artikel <4soarj$f...@freenet-news.carleton.ca>,
dc...@FreeNet.Carleton.CA (Alan Tomalty) you wrote:
>How can we get Nimzo 3?
> Komputer Korner
The worldwide distributor for Nimzo3 is :
Helmut Weigel
Spengergasse 61/17
A - 1050 Wien
Oesterreich / Austria
Fax : ++43 - 1 - 545 10 52
Peter Schreiner
Alan Tomalty
How can we get Nimzo 3?
Komputer Korner
--
Komputer Korner
Seifriz
Hi,
you can order it here if you send email! An English manual is
available now, and the program itself can also be run in English!
I agree about the high speed. On a Pentium 200 MHz it evaluates
70 000 positions per second! Best regards Bert Seifriz
Bruce Moreland
In article <31F020...@europeonline.com>, gambi...@europeonline.com
says...
Is this a typo? That sounds pretty slow, actually.
bruce
--
The opinions expressed in this message are my own personal views
and do not reflect the official views of Microsoft Corporation.