Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: KRRNKRR win in 290: a new record

Author: Dann Corbit

Date: 19:12:55 10/27/05

Go up one level in this thread


On October 27, 2005 at 07:03:18, Yakov Konoval wrote:

>On October 26, 2005 at 20:27:19, Dann Corbit wrote:
>
>>On October 20, 2005 at 10:56:15, Joshua Shriver wrote:
>>
>>>Interesting, a thought on compression though. Why not use bz2? It's just as easy
>>>to implement as zip and has better compression.
>>
>>Bzip2 is horribly slow, both for compressing and decompressing.
>>
>>The extra compression gained is a small percentage.  Fine for web or ftp sites,
>>but (I guess) a bad choice for EGTB compression.
>>
>
>I did some experiments with bzip2-library. It's really slow, but gives much
>better (about 60-80%) compression in comparison with zlib-library.
>
>>7-Zip has lots of algorithms implemented (including bzip2.)  The license is a
>>bit unclear to me.
>>
>
>AFAIK 7-Zip LZMA SDK has a very effective LZMA-algorithm, but using of this
>library is really unclear. Only 2 functions are needed for EGTB-generator-
>compression/decompression of 1 block in the RAM, but there are no such fuctions
>in lzma.dll.
>
>>This looks very interesting (although license is only GPL or commercial):
>>http://www.oberhumer.com/opensource/lzo/
>>
>>What is wanted in a case like EGTB is very fast decompression and very good
>>compression.
>
>The main criteria for generation is very fast compression/decompression, but
>for the final study of generation the best compression is needed.

What I would want from a set of EGTB files is the fastest possible access.
So if decomression of the block was slower than reading the decompressed block,
then the compression is literally "a waste of time."

If you need provably the best compression, then look no further:
http://www.ii.uni.wroc.pl/~inikep/research/pasqda41b.zip

According to this document:
http://www.maximumcompression.com/data/doc.php
pasqda41b achieved 1.3552 bits per byte against a large body of inputs.

Comes with source code.  But expect it to be a slow, slow memory hog.



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.