Opening taxonomy

[hsmyers]

I've become interested in automatic identification of a chess games's
opening. The first solution that occurred was a sledge-hammer sort of
hack where I simply compare the parsed moves of the game with an a
'opening' database. This seems to work well and was fairly simple to
implement. It occurs to me however, that this will fail for
transpositions. It won't likely be a total failure, since the game will
match the 'line' up until the first transposition, but I'd like to do
better. I plan on a kind of backwards position match against the sub-tree
indicated just before the 'fail'. I was wondering if anyone has done
anything similar or would just like to throw in their 2cents worth...

hsm

[Dr. Michael Hale]

You might like to check out gnuchess 4.0pl75, available at
prep.ai.mit.edu in the pub/gnu directory. It has the ability to
identify ECO lines, including transpositions. I can't speak of
how well it does this (anyone else have a comment?), but it may
be a start for you.

Cheers
Mike
--
------------------------------------------------------------------------------
Michael Hale, PhD | EMail: mh...@ac.dal.ca
Asst. Professor | SMail: Medical Physics Dept.
Dept. of Radiation Oncology | Nova Scotia Cancer Centre
Dalhousie University | 5820 University Ave., Halifax, NS
EMail: mh...@ac.dal.ca | B3H 1V7 CANADA
------------------------------------------------------------------------------
"Science has achieved some wonderful things, of course, but I'd far rather be
happy than right any day." - Slartibartfast

[Bookup]

In article <432o7g$f...@mis02.micron.net>, hsmyers <hsm...@micron.net> wrote:
>I've become interested in automatic identification of a chess games's
>opening. The first solution that occurred was a sledge-hammer sort of
>hack where I simply compare the parsed moves of the game with an a
>'opening' database. This seems to work well and was fairly simple to
>implement. It occurs to me however, that this will fail for
>transpositions.

We used to have a utility that gave opening codes to
games based on a BOOKUP 7 database. Each position
had an opening code (whatever code you wanted to use)
and, being BOOKUP, it never missed a transposition.

The real challenge is in the coding system itself.
If a certain position is generally reach half the
time from the English and the rest of the time from
the Queen's Gambit, what "code" does one use?

>It won't likely be a total failure, since the game will
>match the 'line' up until the first transposition, but I'd like to do
>better. I plan on a kind of backwards position match against the sub-tree
>indicated just before the 'fail'. I was wondering if anyone has done
>anything similar or would just like to throw in their 2cents worth...
>
>hsm

Mike Leahy
"The Database Man!"

[hsmyers]

From what I've seen of gnuchess, it does not take kindly to being
high-graded. That is it is more of a case of this piece is connected to
that piece is connected to that piece...and so on. The good news is that
I've already got code that identifies ECO lines. What I'm looking to
write for myself is code that identifies positions. Probably hash
based(as compared to free based<g>) So I guess the subject should really
be positional hashing...

hsm

[hsmyers]

I don't really see that positions that are the same in different lines
are per force a problem. Actually this is a significant piece of
information. Consider that opening books that condemn a position in one
book and praise it in another. I'd like to spot situations like
that...perhaps some system where you look for further matches in the
process, having found an initial match.

hsm

[Michel Behna]

In article <432o7g$f...@mis02.micron.net> hsmyers <hsm...@micron.net> writes:
>From: hsmyers <hsm...@micron.net>
>Subject: Opening taxonomy
>Date: 12 Sep 1995 01:36:48 GMT

>I've become interested in automatic identification of a chess games's
>opening. The first solution that occurred was a sledge-hammer sort of
>hack where I simply compare the parsed moves of the game with an a
>'opening' database. This seems to work well and was fairly simple to
>implement. It occurs to me however, that this will fail for

>transpositions. It won't likely be a total failure, since the game will

>match the 'line' up until the first transposition, but I'd like to do
>better. I plan on a kind of backwards position match against the sub-tree
>indicated just before the 'fail'. I was wondering if anyone has done
>anything similar or would just like to throw in their 2cents worth...

>hsm

Here's an inetresting idea: has anyone cared that games that end up
transposing to a particular opening pass thru one or more openings. Is there
anything out there that would classify them and show at what moves they were
considered to be what ECO code? Just an idea.

Michel Behna - mbe...@promus.com
DoD #1821 - Suzuki Katana 600