Re: bayes issue - any ideas?

[sub]

Rod,

good idea.

I should have mentioned - the upgrade was over a month ago - at which time I migrated the bayes db.

Perhaps I can export/import the bayes db, and see if it changes the outcome.

my gut feel is the even if the spam db is different, the new spamassassin/bayes behaves differently

closer inspection of the bayes debug shows this in the new one

'Grant' => 0.0319406228099444
'Pam' => 0.0455135183600126
'bret' => 0.0489090909090909
'Bret' => 0.0489090909090909
'ebony' => 0.942158296767548
'Alex' => 0.0634177360303953
'slut' => 0.927820169651628
'alex' => 0.0741859079412405
'marc' => 0.0792365757032377
'michaels' => 0.0810433553800303

and in the older one:

'Alex' => 0.999279251170047
'slut' => 0.990941176470588
'Grant' => 0.990941176470588

a short analysis:
so they both think 'slut' is spammy,
they have vary opposing opinions of the words grant/alex,
and words like michaels doesnt even appear on the old debug.

Perhaps the use of all the names has poisoned the bayse analysis. The name might be ligit users, and so spam learning says these are ham words.

Perhaps the new version is more sensitive to a few ham words, and is attempting to do less false positives.

Scott
[EMAIL PROTECTED]

>
> Hi,
>
> Export your old data and import it into the new database.
>
> man sa-learn.
>
> Regards,
>
> Rick
>

Rick Macdougall <[EMAIL PROTECTED]> wrote on 07/10/2005 10:54:40 AM:

> [EMAIL PROTECTED] wrote:
>
> >
> > Hi,
> >
> > (please to reply to [EMAIL PROTECTED] as well as the list
> > please)
> >
> > I recently upgraded and now I am getting spam through, with vary
> > obvious rude words in the subject and body. I have narrowed down the
> > difference in bayes results.
> >
> > I am now running
> > SpamAssassin version 3.0.4
> >   running on Perl version 5.6.1
> >
> > My old server (which is still around to test)
> > SpamAssassin version 2.63
> >
> > all via Mailscanner
> >
> > The new spamassassin gives 0/5 for bayes, and the old one gives 5.4/5
> > for bayes
> >
> > here is a bit of the email (I know its not the raw original, but good
> > enough for now for bayes)
> >
> > Subject: Stupid Brunette Amateur Undresses & Fucks
> > An illustrated story from Grant
> > Ebony slut gets stuffed by a monster shlong
> > A brunette mom doing a black man in pics and movies from Marc
> > Pam Anderson & Bret Michaels action movies from Monster
> > http://www.pynojosyhu.com?SVifSQ,jWgfSRVW.UbhjPRQ,hVX,jP
> > A shemale nurse giving head to a black man from Mistress
> > A non nude poser from Alex
> > Fourty three yo prostitute fuck
> >
> >
> > Bayes debug from new server :SpamAssassin version 3.0.4
> >
> >
> > debug: bayes token 'Anderson' => 0.000156805596036141
> > debug: bayes token 'UD:UbhjPRQ,hVX,jP' => 0.999833872707659
> > debug: bayes token 'SVifSQ,jWgfSRVW.UbhjPRQ,hVX,jP' => 0.999816739389131
> > debug: bayes token 'SVifSQjWgfSRVWUbhjPRQhVXjP' => 0.999816739389131
> > debug: bayes token 'sk:svifsq,' => 0.999816739389131
> > debug: bayes token 'sk:svifsq' => 0.999816739389131
> > debug: bayes token 'UD:SVifSQ,jWgfSRVW.UbhjPRQ,hVX,jP' =>
> > 0.999816739389131
> > debug: bayes token 'svifsqjwgfsrvwubhjprqhvxjp' => 0.999816739389131
> > debug: bayes token 'sk:SVifSQ,' => 0.999816739389131
> > debug: bayes token 'svifsq,jwgfsrvw.ubhjprq,hvx,jp' => 0.999816739389131
> > debug: bayes token 'sk:SVifSQ' => 0.999816739389131
> > debug: bayes token 'shemale' => 0.996181818181818
> > debug: bayes token 'mistress' => 0.993492957746479
> > debug: bayes token 'Fucks' => 0.992426229508197
> > debug: bayes token 'Amateur' => 0.988790846768044
> > debug: bayes token 'pam' => 0.0215564263263569
> > debug: bayes token 'Mistress' => 0.978
> > debug: bayes token 'Brunette' => 0.976426848800704
> > debug: bayes token 'anderson' => 0.0240121331183822
> > debug: bayes token 'Grant' => 0.0319406228099444
> > debug: bayes token 'Michaels' => 0.0380548617527754
> > debug: bayes token 'amateur' => 0.958828640105853
> > debug: bayes token 'prostitute' => 0.958
> > debug: bayes token 'Pam' => 0.0455135183600126
> > debug: bayes token 'bret' => 0.0489090909090909
> > debug: bayes token 'Bret' => 0.0489090909090909
> > debug: bayes token 'ebony' => 0.942158296767548
> > debug: bayes token 'Alex' => 0.0634177360303953
> > debug: bayes token 'slut' => 0.927820169651628
> > debug: bayes token 'alex' => 0.0741859079412405
> > debug: bayes token 'marc' => 0.0792365757032377
> > debug: bayes token 'michaels' => 0.0810433553800303
> > debug: bayes token 'fucks' => 0.908054708334054
> > debug: bayes token 'brunette' => 0.903431604883336
> > debug: bayes token 'Marc' => 0.0978953886133904
> > debug: bayes token 'nude' => 0.89705901511311
> > debug: bayes token 'grant' => 0.113569258541575
> > debug: bayes: score = 0.519120285425019
> >
> >
> > bayes debug from older server:SpamAssassin version 2.63
> > debug: bayes token 'mom' => 0.999800086542622
> > debug: bayes token 'Alex' => 0.999279251170047
> > debug: bayes token 'Brunette' => 0.998683760683761
> > debug: bayes token 'nude' => 0.998560747663551
> > debug: bayes token 'brunette' => 0.998560747663551
> > debug: bayes token 'Fucks' => 0.998159362549801
> > debug: bayes token 'stupid' => 0.997909502262443
> > debug: bayes token 'Ebony' => 0.993492957746479
> > debug: bayes token 'slut' => 0.990941176470588
> > debug: bayes token 'Grant' => 0.990941176470588
> > debug: bayes token 'non' => 0.989881305839074
> > debug: bayes token 'nurse' => 0.985096774193548
> > debug: bayes token 'sk:SVifSQ,' => 0.985096774193548
> > debug: bayes token 'monster' => 0.985096774193548
> > debug: bayes token 'UD:UbhjPRQ,hVX,jP' => 0.985096774193548
> > debug: bayes token 'Mistress' => 0.978
> > debug: bayes token 'Pam' => 0.978
> > debug: bayes token 'Amateur' => 0.975545163594609
> > debug: bayes token 'Undresses' => 0.958
> > debug: bayes token 'poser' => 0.958
> > debug: bayes token 'Monster' => 0.958
> > debug: bayes token 'fuck' => 0.90580008624735
> > debug: bayes: score = 1
> >
> > Any ideas?
> >
> > Scott
> > [EMAIL PROTECTED]