Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Another opening book question

Author: Dann Corbit

Date: 11:59:57 07/17/02

Go up one level in this thread


On July 17, 2002 at 14:36:29, Dieter Buerssner wrote:
[snip]
>Did you check, if this is worth the work? Do you get now and then opening moves
>by the probing of the mirrored position? How often? Once in 10 games, in 100
>games? ...

It depends to a great extent on the density of your database.

If you have 3 positions, you will not find many hits.

If you have 30 million positions, you will find lots of them.

From some percursory analysis of the CAP project data, I think that once you
have a million positions in your database, you will get about one transform hit
per hundred thousand, and then it will increase steadily as you add more
positions.

I base that on my initial work with the project.  My first build had the
database not case sensitive.  I had a unique clustered index on the EPD position
with ignore duplicates.  As I added more and more data, I started to see a few
rejections in the first few hundred thousand rows.  As it grew over a million, I
had so many hits I investigated to find that I was getting rejected rows where
the pieces were all identical, but some colors of the pieces were different.
Obviously, I had to make a new database that was case sensitive and reload the
data.

Sample where a case insensitive server would reject the transaction with
"identical key":
[D]3k4/5n2/8/8/1n6/5N2/8/3K4 w - -
[D]3k4/5n2/8/8/1N6/5N2/8/3K4 w - -

If you have hundreds of millions of positions in your database, you will get a
huge number of successful transformations.

If you have a very small database, it is a waste of time.

IMO-YMMV



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.