Newsgroups: comp.compression
Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!sarah!bingnews!kym
From: kym@bingvaxu.cc.binghamton.edu (R. Kym Horsell)
Subject: Re: theoretical compression factor
Message-ID: <1991Apr2.224833.7907@bingvaxu.cc.binghamton.edu>
Organization: State University of New York at Binghamton
References: <20144@alice.att.com> <1991Apr2.034441.28170@bingvaxu.cc.binghamton.edu> <20159@alice.att.com>
Date: Tue, 2 Apr 1991 22:48:33 GMT

In article <20159@alice.att.com> jj@alice.UUCP (jj, like it or not) writes:
[I wrote]
>>Is p lg p a hard-and-fast bound or not? I still think it's an average.

>It's ALWAYS the average rate.  Given that you're using compression,
>you can't measure any other thing, unless you have a fixed message
>set, and that's all.  (In which case you should enumerate your
>messages, perhaps huffmanwise...)  If you have  any set of data
>with the overall 'p' known, that is the average rate for THE WHOLE
>SET. Parts of it may do better, because they may hit shorter
>symbols (for instance), but if you look at the probabilities
>of THAT PART, you should find that you don't do better than
>the p log p chosen over your locality, without a model.

Ta.

-kym