info@concurrent.co.uk (Information Server) (06/04/91)
In article <817@spam.ua.oz> ross@spam.ua.oz.au (Ross Williams) writes:
[...]
SAKDC is at this stage not very accessible or portable.
[...]
As far as I am aware, SAKDC gives the best compression of ANY
algorithm in existence (if you can do better - scream out!). Evidence
for this can be found in my book and the Bell book. Table B-1 of the
Bell book gives the compression performance on the Calgary (Bell)
corpus for 18 algorithms. The best compression is given by the Markov
algorithms of which PPMC is the best, yielding 2.48 bits per symbol
(the next closest is 2.74). From the O20000 column of Table 39 of my
book (and omitting the files paper3..paper6 as Bell Cleary and Witten
did) SAKDC yielded 2.464 bits per symbol. This was using a large
amount of memory. For a small amount of memory (200 nodes), SAKDC
yields about 18% (absolute) better compression than PPMC (3.344b/s vs
4.784b/s (includes paper3-paper6)).
[...]
Are there any freely available implementations of PPM, PPMC or any other
Markov algorithm compressors?
John Connett
jrc@concurrent.co.uk