[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Ifile-discuss] A few questions about ifile features compared to oth
From: |
Jason Rennie |
Subject: |
Re: [Ifile-discuss] A few questions about ifile features compared to other classifiers |
Date: |
Fri, 31 Oct 2003 09:04:46 -0500 |
address@hidden said:
> 1) storage based on a real database ( Berkeley DB for instance)
> instead of your file format ? Do you think it would improve its
> performance ?
ifile is pretty fast as is. Also, it's nice to have the database in ASCII
format---makes it easier for people to figure out what is going on and it
also makes the code easier to debug when there are problems. Another
nice benefit is that ifile is small and totally standalone, which has
its benefits.
address@hidden said:
> Have you tried to adapt the code writed in bogofilter for instance to
> add such features to ifile ? Do you think It's worth trying (I'm
> volunteering) ?
It is worth trying. I think the most useful thing to do would be to
adapt features from POPFile, since that would greatly improve ifile's
spam filter capabilities.
I personally use bogofilter for spam filtering and ifile for everything
else.
address@hidden said:
> 3) A last thing about sort accuracy. I read in one page that some of
> you reached 96% accurate classification. How have you calculated that
> ? Do you all have such high percentage ?
My accuracy has always been in the 80-90% range, though I keep a lot of
folders (currently 86), so it might be a bit higher for someone with
fewer folders.
ifile marks each e-mail with an X-filter: header that identifies the
version number and the folder into which it classified the e-mail.
Also, the mh-ifile scripts automatically update an ~/.idata_accuracy
file, which simply calculates (filters-refiles)/(filters), i.e. (# e-mails
classified correctly)/(total # e-mails classified).
Jason
[Prev in Thread] |
Current Thread |
[Next in Thread] |
- Re: [Ifile-discuss] A few questions about ifile features compared to other classifiers,
Jason Rennie <=