Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: WHAT !!!!! NO WAY (thats about 10bits/position) HOWWWWWW????

Author: Dann Corbit

Date: 14:17:32 01/04/02

Go up one level in this thread


On January 04, 2002 at 17:10:42, Nino wrote:
>On January 04, 2002 at 05:14:16, Dann Corbit wrote:
>>Suppose that we want to make a tablebase file that is not quite so good as a
>>Nalimov tablebase file, but it still gives us a lot of information and is very
>>compact.  Let's consider this winning position
>>[D]r6r/8/8/3k4/8/3K4/6R1/4QR2 w - - acn 2125; acs 0; bm Qe4+; ce 32760; dm 4; pv
>>Qe4+ Kc5 Rc2+ Kb5 Rb1+ Ka5 Ra2#;
>>
>>Now, as anyone knows, the queen being a sliding piece with a lot of available
>>options, any of these positions will result in the same sort of direct mate in
>>4, because we will arrive at the same position, or a mirror of it:
>>r6r/8/8/3k4/8/3KQ3/6R1/5R2 w - - ce 32760; pv Qe4+;
>>r6r/8/8/3k4/8/3K4/4Q1R1/5R2 w - - ce 32760; pv Qe4+;
>>r6r/8/8/3k4/8/3K4/6R1/4QR2 w - - ce 32760; pv Qe4+;
>>r6r/8/8/3k4/5Q2/3K4/6R1/5R2 w - - ce 32760; pv Qe4+;
>>r6r/8/8/3k4/6Q1/3K4/6R1/5R2 w - - ce 32760; pv Qe4+;
>>r6r/8/8/3k4/7Q/3K4/6R1/5R2 w - - ce 32760; pv Qe4+;
>>r6r/8/8/3k4/1Q6/3K4/6R1/5R2 w - - ce 32760; pv Qe4+;
>>r6r/8/8/3k4/Q7/3K4/6R1/5R2 w - - ce 32760; pv Qe4+;
>>r6r/8/8/4k3/8/3QK3/1R6/2R5 w - - ce 32760; pv Qd4+;
>>r6r/8/8/4k3/8/4K3/1R1Q4/2R5 w - - ce 32760; pv Qd4+;
>>r6r/8/8/4k3/8/4K3/1R6/2RQ4 w - - ce 32760; pv Qd4+;
>>r6r/8/8/4k3/6Q1/4K3/1R6/2R5 w - - ce 32760; pv Qd4+;
>>r6r/8/8/4k3/7Q/4K3/1R6/2R5 w - - ce 32760; pv Qd4+;
>>r6r/8/8/4k3/2Q5/4K3/1R6/2R5 w - - ce 32760; pv Qd4+;
>>r6r/8/8/4k3/1Q6/4K3/1R6/2R5 w - - ce 32760; pv Qd4+;
>>r6r/8/8/4k3/Q7/4K3/1R6/2R5 w - - ce 32760; pv Qd4+;
>>5r2/6r1/3kq3/8/3K4/8/8/R6R b - - ce 32760; pv Qe5+;
>>5r2/4q1r1/3k4/8/3K4/8/8/R6R b - - ce 32760; pv Qe5+;
>>4qr2/6r1/3k4/8/3K4/8/8/R6R b - - ce 32760; pv Qe5+;
>>5r2/6r1/3k4/5q2/3K4/8/8/R6R b - - ce 32760; pv Qe5+;
>>5r2/6r1/3k4/6q1/3K4/8/8/R6R b - - ce 32760; pv Qe5+;
>>5r2/6r1/3k4/7q/3K4/8/8/R6R b - - ce 32760; pv Qe5+;
>>5r2/6r1/3k4/1q6/3K4/8/8/R6R b - - ce 32760; pv Qe5+;
>>5r2/6r1/3k4/q7/3K4/8/8/R6R b - - ce 32760; pv Qe5+;
>>2r5/1r6/3qk3/8/4K3/8/8/R6R b - - ce 32760; pv Qd5+;
>>2r5/1r1q4/4k3/8/4K3/8/8/R6R b - - ce 32760; pv Qd5+;
>>2rq4/1r6/4k3/8/4K3/8/8/R6R b - - ce 32760; pv Qd5+;
>>2r5/1r6/4k3/6q1/4K3/8/8/R6R b - - ce 32760; pv Qd5+;
>>2r5/1r6/4k3/7q/4K3/8/8/R6R b - - ce 32760; pv Qd5+;
>>2r5/1r6/4k3/2q5/4K3/8/8/R6R b - - ce 32760; pv Qd5+;
>>2r5/1r6/4k3/1q6/4K3/8/8/R6R b - - ce 32760; pv Qd5+;
>>2r5/1r6/4k3/q7/4K3/8/8/R6R b - - ce 32760; pv Qd5+;
>>
>>Now, if we take these positions, and lexically sort them, we can store the
>>minimum one into a database and be able to generate all the others from it.
>>
>>Then, if we have some position and want to see if it is stored in the database,
>>we could simply perform the same transformation and lookup the lexically
>>smallest entry to see if it exists.  We know that for any analysis our database
>>contains, the best move will be at least as good as the value stored (there
>>might be a better move, but the selected move will be "At Least" as good as the
>>presented analysis.
>>
>>As you can see, this would also allow partial database files, and it will allow
>>their use before they are completed.
>>
>>The data can therefore be remarkably compressed in some instances.  I have no
>>figures for averages, as I have not carefully studied it yet.  But in the
>>position above, a single EPD string maps to 32 answers.  Therefore the net
>>storage needed for that particular position is 1/32 times the size of a binary
>>EPD record.
>>
>
>Dan how can this be possible?? You cant store all that information in so few
>bits.  What you are saying basically is impossible.  Computer 101 says that a
>bit can only have 2 states ...etc etc.  Any other commnets from members out
>here?  Since I dont see any they are probably laughing about this "NEW IDEA"
>
>Perhaps it is time for you to explain this in a bit more detail or am I the only
>one who does not understand.

It gets even stranger.  Suppose that you have 100 positions that result from
that 64 bit key.  Each position will require less than a single bit on average.

The information is fuzzy information, in that it is not perfect.  There might be
a shorter mate (for instance).  But one single key can hold all the information
needed for a whole cluster of positions.  One of those things to file under
"strange but true"  -- have you looked at the folder in the FTP site here:

>>Here is some information on the idea, which came from Les Fernandez:
>>ftp://cap.connx.com/pub/les/Least/

In particular, this file:
ftp://cap.connx.com/pub/les/Least/egtbc.ppt

is a powerpoint presentation that explains how it works.  Furthermore, the VB
application with source code in that same folder is a *working* demonstration of
the idea.  He has not added a filter to remove illegal moves (I am just using
KKE for that right now) but you can easily see how it works.

>>A very nice thing about this sort of database is that it is fairly easy to
>>produce partial files with lots of pieces in them.



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.