Re: Spam filter reviews?
On 18 Mar 2003, Bill Wohler wrote:
> Anthony Campbell <ac@acampbell.org.uk> writes:
>
> > I installed bogofilter about 10 days ago and have been extremely
> > impressed. Previously with spamassassin I was getting several
> > false-negatives daily but now I hardly ever get even one. The training
> > scheme seems to be very effective. The same applies to false-positives;
> > there were a few to start with but those, too, have now been eliminated.
>
> Thanks Anthony. That's just what I was looking for.
>
> A couple of interesting data points from learning on my corpus which
> contains 90,000 ham and 200 spam messages. The spambox just got
> cleaned out unfortunately, but just as unfortunate, I'll have a couple
> of thousand in a week or two so training will be quick:
>
> bogofilter took 52 minutes to build:
>
> -rw-r--r-- 1 wohler users 73723904 2003-03-18 08:46 goodlist.db
> -rw-r--r-- 1 wohler users 581632 2003-03-18 08:46 spamlist.db
>
> spamprobe took 297 minutes to build:
>
> -rw------- 1 wohler users 484311040 2003-03-17 22:36 sp_words
>
> sa-learn (--no-build on each folder followed by a single --rebuild)
> took 154 minutes to build:
>
> -rw------- 1 wohler users 157448 2003-03-18 16:27 bayes_journal
> -rw------- 1 wohler users 707 2003-03-18 16:27 bayes_msgcount
> -rw------- 1 wohler users 10264576 2003-03-18 16:27 bayes_seen
> -rw------- 1 wohler users 82419712 2003-03-18 16:27 bayes_toks
>
> In addition to your observations, bogofilter also wins on the space
> and time angles..
>
I tried sa-learn previously and didn't have much luck. Even though I
gave it over 2000 Ham emails to learn and about 700 Spam it still
didn't seem to be doing much. I continue to be happy with bogofilter.
AC
--
ac@acampbell.org.uk || http://www.acampbell.org.uk
using Linux GNU/Debian || for book reviews, electronic
Windows-free zone || books and skeptical articles
Reply to: