On Mon, 25 Feb 2008 16:32:53 +0100
"Olivier Wulveryck" <[EMAIL PROTECTED]> wrote:

> Hello everybody,
> 
> I have a problem with my dspam installation.
> 
> Here is what I want to do:
> 
> Let's say that my MX is mailsystem.com and that I have a test user
> [EMAIL PROTECTED]
> The MTA (postfix) is hosted on a machine mta.mydomain.com
> The MDA (cyrus) is hosted on a machine mda.mydomain.com and
> communicates with postfix via lmtp.
> 
> What I wanted to do basically was something like this:
> 
> SMTP                     lmtp                     lmtp
> -----> MTA.mydomain.com -----> DSPAM.mydomain.com ----->
> MDA.mydomain.com

OK, so delivery to the cyrus works.

> DSPAM is using a mysql backend and postfix is using DSPAM as a
> content_filter:
> content_filter = lmtp:[dspam.mydomain.com]:24

OK.

> 
> What I could see with dspam_stats -H [EMAIL PROTECTED] was something
> like this
> 
> [EMAIL PROTECTED]:
>               TP True Positives:              0
>               TN True Negatives:              100
>               FP False Positives:             0
>               FN False Negatives:             0
>               SC Spam Corpusfed:              0
>               NC Nonspam Corpusfed:           0
>               TL Training Left:            2400
>               SHR Spam Hit Rate         100.00%
>               HSR Ham Strike Rate:        0.00%
>               OCA Overall Accuracy:     100.00%
>         
> The point is that no spam were detected, and I'm sure there was...

Blank database, no training so it's normal.

> So I decided to do a training of dspam this way:
> I've downloaded 20021010_spam.tar.bz2 and 20030228_easy_ham.tar.bz2
> from http://spamassassin.apache.org/publiccorpus/

This are very old; very probably the spams look different these days.

> untar, and the run
> 
> # dspam_train [EMAIL PROTECTED] spam/ easy_ham/
> 
> every single spam was failed and detected as innocent e.g: 
> [test: spam   ] 00013.d3f0b591a65f116ea5d9d4ad91 result: FAIL
> (Innocent) [fn] Subject: Get the Child Support You Deserve
> 11.180

Umm, this is the output from what ?

> Retraining didn't help, and dspam_stats gave me the following results:
> 
> [EMAIL PROTECTED]:
>               TP True Positives:              0
>               TN True Negatives:             65
>               FP False Positives:             0
>               FN False Negatives:            14
>               SC Spam Corpusfed:              0
>               NC Nonspam Corpusfed:           0
>               TL Training Left:            2435
>               SHR Spam Hit Rate           0.00%
>               HSR Ham Strike Rate:        0.00%
>               OCA Overall Accuracy:      82.28%

What I don't see here is the corpus training. I see 65 non-spam mails
correctly identified as ham and 14 identified wrong.

> I was getting mad so I've decided to change my configuration in a more
> standard one, in order to reinject the mails back to postfix via SMTP.
> Of course the result is the same.

Yeh, of course.

> I may have missed something... Is there something special with virtual
> users or something. I've already RTFM or STFW ;-)

Yeh, maybe, please read /usr/local/share/doc/dspam/README.FreeBSD and
pass _all_ the info requested there. I wrote it on purpose, you know ;-) 

Run dspam with debug and pass the output.

> Don't think this have any importance but dspam is hosted on FreeBSD.

Except that we know mail/dspam works very well on FreeBSD, no, there


-- 
IOnut - Un^d^dregistered ;) FreeBSD "user"
  "Intellectual Property" is   nowhere near as valuable   as "Intellect"

Attachment: signature.asc
Description: PGP signature

Reply via email to