Bug#179804: PPM compression algorithms. (fwd)
close 179804
thanks
---------- Forwarded message ----------
Date: Thu, 13 Feb 2003 13:05:13 -0600 (CST)
From: Drew Scott Daniels <umdanie8@cc.UManitoba.CA>
To: Magnus Ekdahl <maguno@ludd.luth.se>
Subject: Re: PPM compression algorithms.
Thank you for the detailed replies and for the analysis.
Can you close 179804 with the appropriate information from you and Charles
Bloom? I haven't learned the BTS system well enough. I assume Charles
Bloom's comments are a sufficient reason to close bug 179804.
Drew Daniels
---
Ok, I'll do that.
/Magnus Ekdahl
---
On Thu, 13 Feb 2003, Magnus Ekdahl wrote:
> On Thu, 13 Feb 2003, Drew Scott Daniels wrote:
>
> > I'm cc'ing this packaging discussion to Charles Bloom.
...
> > I see that you say that PPMd is "Marginally better than PPMZ" and that
> > it's better than "PPMZ(2)". Does that mean that you got PPMZ2 v0.8
> > from May 14, 2002 working under linux or did you do a comparison with
> > older versions or under MS Windows?
>
> Originally I was using an old version of PPMZ when doing the statement
> that PPMd is "Marginally better than PPMZ". But since you brought it to my
> attention again I tested it again with the newest version of the linux
> port. The results from the test can be seen in
> /usr/share/doc/ppmd/PPMZ.txt in ppmd-9.1-2 (attached).
---
>From Charles Blom:
----
Hi guys, a few notes :
PPMZ was never really intended for direct use in production software;
it's more of a test-bed. For example, it was never really optimized.
Malcolm Taylor took a lot of the ideas of PPMZ and put them in RKive,
which is very fast. Also, the so-called "PPMd" and PPMii are based on
PPMZ, and are more finely tweaked and optimized.
PPMii is very good; it's similar to PPMZ but tweaked and improved in
various ways. The biggest difference is that he is doing a bit of
"Context mixing and blending" across contexts which are not directly
linked in the Markov chain.
PPMd - I don't think there are any actual papers on. This is NOT the
"PPMD" algorithm of Bill Teahan, so it's somewhat mis-named. I believe
it's very similar to PPMZ, but with some tweaks and special custom support
for text files and binaries.
>Charles Bloom: Do you know why PPMd is performing better? Do you have any
>suggestions for code that would be useful for me? I'd like to license my
>code later like Ogg, gzip, zip and bzip2 so that it can be used in
>commercial applications. Are you still developing PPMZ(2) or any other
>compressors?
I'm no longer actively working in compression; I haven't really touched
the
PPMZ's in several years. In modern stuff, PPMd is obviously good. The
primary algorithms in this area are PPMZ and PPMii ; there are
implementations
like BOA and PPMonstr that are very good, and many more.
-------------------------------------------------------
Charles Bloom cb@cbloom.com http://www.cbloom.com
Reply to: