Re: Fwd: Re: ifile vs. Graham modified naive bayes?

'bout time we created separate "stable" and "development" branches :)

[email protected] said:
> -> On the one hand, performance on a large corpus gets pretty painful.
>    All the word stats get read through, whether they get used or not. 

[email protected] said:
> -> It would also be nice to do more sophisticated "symbol tagging." 

These are both very good ideas.

Better handling multi-part messages would also be useful.  I find that
attached images tend to confuse ifile.  It breaks the image into thousands
of two- or three-character strings.  The sheer number of "words" that an
image produces overpowers the real content of the message.  Nicer would 
be to identify the whole attachment as an image and generate an image 
"word".

Any volunteers? :)

Jason