Stocks and P-R-O-F-I-T

[Max Kipness]

> Hello,
> 
> I had recently tried what I thought was a good technique, and created
a
> script that fed all email from every MailScanner white listed email
> address into sa-learn as ham nightly, without doing a check on the
> emails. This was obviously a bad choice as jokes and other spam like
> emails must have processed for months.
> 
> Anyway, I scrapped the bayes database and started from scratch using
the
> a sample bayes db from FSL (I think it's called). From there I've been
> feeding quite a bit of spam into sa-learn for about a week or two. I'd
> say I've fed about 400 spam mails thus far. However, as of today I'm
> still getting the p-r-o-f-i-t and stock spasm with bayes scores of
> anywhere from 10% to 50%.

>>My question is what kind of stock spams are they?

>>Are they image based, or text based?

>>If it's image, bayes won't help you much, as bayes doesn't understand
>>images.

> My question is how long or how many emails should it take bayes to
> figure out these spam emails? Is there a way of viewing the progress?
> With the other scores from DCC, Pyzor, Razor, the score is close to
> being tagged as spam, but sometimes it's not quite there because of
the
> bayes score.

>>for image spams, try adding the SARE stocks ruleset.

Yes, they are partially or sometimes full images.

I have the SARE Stock ruleset installed, the problem is that whether or
not this ruleset is being triggered, bayes is sometimes giving a
negative score. Are you saying bayes cannot be trained to score high on
messages that have images? I would think it would examine the fact that
it's an image and header information, but maybe I'm wrong.


Thanks,
Max