My BSD wrote:
> On Tue, 05 Sep 2006 10:17:21 +0800
> W B Hacker <wbh@???> wrote:
>
>
>> ...
>>
>>Went and had a look at bsfilter. Have some questions:
>>
>>- '...simpler...' Possibly. But Bayesian ONLY?
>>
>>Is it as *effective* as SA (with or without SA's Spam_Bayes?)
>>
>
> Although it's been a while since I used SA, I would say yes -- as long as
> all you require is a solid classification engine with no bells and whistles.
>
>>- '...not a resource hog' Perhaps not - if it is doing far fewer
>>comparisons.
>>
>
> My first experience with SA was by way of the Mailscanner wrapper, the
> combination of which made me end up with a whole pig pen as far as resources
> were concerned. It became a little better, but still unacceptable after
> running SA directly.
Surprised - not that we consider it 'sparing' of resources, but run directly at
SMTP time, and then only when needed, SA has behaved itself reasonably well.
Woudl grant that if run on 100% of offered traffic instead of 11%, thta woudl
chage .. but presume you also don't call bsfilter on all arrivals, either.
Given that we have Spam_Bayes optioned-off, we are using diametrically opposite
approaches, hence my curiosity.
>
>
>>But does it remain so as the Bayes DB grows? Does it become more/less
>>effective with such growth? i.e - begin to learn bad habits.
>>
>
> So far, no problems. The beauty of bsfilter is that it can be set up to
> not train with every scan, instead you train it on classification errors
> (and they are few and far in between). Also, it prunes itself when the
> databases get bigger than the set treshold.
>
Sounds like a good balance. We had caught Spam_Bayes going contrary a couple of
times - i.e. 'blessing' spam instead of condemning it.
>
>>ISTR that Ruby is generally slower in execution than comparable perl,
>>syntax elegance, security issues or lack therof entirely aside.
>>
>
> Don't know the answer. All I can say is that I am extremely satisfied
> with the way it works for me.
>
>
>>my curretn impression is that bsfilter doesn't seem to be nearly as
>>tailorable, nor as universal a scanning solution as SpamAssassin.
>>
>>Any evidence otherwise? Comparisons?
>>
>
> Please see my first reply above. I would suggest that you withhold judgment
> (and pontification) until you have tested it for yourself. :)
I'm not overly fond of the way SA must be tailored - but have come to rely on
being able to change the weighting of individual tests, add new ones, omit
others. Fairly 'fine-grain' even if a bit of a PITA.
I'd like something 'lighter' resource-load-wise, have tried several others
(DSpam & Bogofilter most recently), but do want to keep smtp-time scanning.
So far, SA still serves our needs as well as any other, and is well understood.
Bayes, for our needs anyway, seems to work best at the per-user level, in a
decent MUA, so will give bsfilter a miss for now.
Thanks for sharing your experience with it. I had no prior knowledge it even
existed until you posted.
Regards,
Bill
>
>
>>Bill
>>
>
>
> --
> My
>
