Re: Select Announment

Mikael Ylikoski <[email protected]>:

> It is similar to ifile, but is more configurable.
> Some differences from ifile are:
> - Client/server model
>   There is a daemon that does all classification. Clients communicate
>   with the daemon through a local socket.
> - Plug-in based classification algorithms Classifiers are
>   dynamically loaded. Some of the available classifiers are Naive
>   Bayes, Rocchio, Perceptron, "sender" and "thread" classifiers.
> - Language dependent processing
>   Can choose different locale, stopwords and stemming algorithms
>   depending on the language of the mail. A built-in language identifier
>   can guess the language if none is specified.

> http://www.student.nada.kth.se/~d97-miy/select/

this sounds _very_ interesting!  i went to the site, and even the
links are interesting, at least for me.

> This version of Select is not intended for practical use.

... which is what bothers me.  frankly, i fought spam for ages until i
came across ifile, and i'm ever grateful for its existence.  i
wouldn't want to compile a complicated set of C++ programs just to
benchmark it against ifile, because i use ifile only for spam-
detection, which it does very good, although this could be considered
an insult to ifile, which is also a good general classifier for any
email type message.

> It was originally written for research purposes and is missing
> essential features. Furthermore I don't claim that all the features
> that it does have are useful in practice. For example, the language
> dependent processing did not do much difference in my tests.

ok.  there are things missing from ifile many wish for, like decoding
the common transport-encodings of messages for the many spams avoiding
detection by base64'ing themselves.  is this easy with select, i mean,
is there a module for this?

> Notable deficiencies in this version are:
> - Saving and loading of databases is not fully implemented.
> - The clients, except for the testing program, are very primitive.
> - There is no feature selection.

depending on the modularity and programming style and documentation
(you know what i mean), this could be easily improved.  on a scale
{bad, medium, good}, what would you call the commentation of your
code?

> This is not an attempt to create a competitor to ifile. I merely
> release this in case someone might find some use for it. I don't
> know if I will continue developing it.

again, this bothers me a little.  how about really writing a simple
application based on `select', so i can pipe messages through it and
have it determine if it is spam?  i think this would make many people
"hungry", although applying select to mere spam-detection might be
underestimating its capabilities.

clemens

------------------------ Yahoo! Groups Sponsor ---------------------~-->
Get 128 Bit SSL Encryption!
http://us.click.yahoo.com/JjlUgA/vN2EAA/kG8FAA/CefplB/TM
---------------------------------------------------------------------~->

To UNSUBSCRIBE, send an empty message to
[email protected]

ifile web site: http://www.ai.mit.edu/~jrennie/ifile/ 

Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/