Massive Scoring differences in Versions
bofh <[email protected]> Fri, 28 Jun 2013 12:20:17 +0200 (CEST)
| Newsgroups | gmane.mail.spam.dspam.devel |
|---|---|
| Message-ID | <[email protected]> |
--===============4912529444302958014==
Content-Type: multipart/alternative;
boundary="----=_Part_228420_103341265.1372414817211"
------=_Part_228420_103341265.1372414817211
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
Hello everyone,
After doing some testing in my Production system :-( i found something intresting.
I Use now OSB again, TOE, and the other standards together with Mysql, imple,mented by amavisd - Zimbra setup
So now if i use latest git Version i get Results like:
X-DSPAM-Result: Spam
X-DSPAM-Processed: Fri Jun 28 04:14:52 2013
X-DSPAM-Confidence: 0.9076
X-DSPAM-Probability: 1.0000
X-DSPAM-Signature: 51ccf19c82621724820715
Which seems odd, i send the same test message from gmail over and over again.
reclassify it retrain it - training give PASS, still next time same thing again
X-DSPAM-Result: Says spam, probability always 1.0000
Then i testet 3.9.0 - because iam using this on zimbra 3.9.0 is the default installation (i jsut symlinked the git version so i can switch quickly)
The difference was massive out of the box
without retrain the last message i got
X-DSPAM-Result: Innocent
X-DSPAM-Processed: Fri Jun 28 04:30:03 2013
X-DSPAM-Confidence: 1.0000
X-DSPAM-Probability: 0.0023
X-DSPAM-Signature: 51ccf52b148228935421196
Same Message same Sender just about 30 seconds later after switched dspam Version.
only difference was dspam 3.9.0 instead of git.
Well i dont say of course this is a better result, i just say its looks more like what ife expected after retrain it 20 times as innocent.
Of course i checked if iam using the right database (dspam stats and so show me the right stats, because i switched to mysql theres no hashfiel anywhere so i think both version should really work with the same database)
also as a side note, there was almost no difference with feature tb - off and tb=9 in the git version
btw why is it marked as spam anyway with a confidence of 0.9 it should be
my guess its because of probability is always 1.00000 , maybe something off there
So after some Digging in the History i see something about changed calculation in OSB and markov weighting - well exactly waht iam using
and yes there is a changed calculation method -
i just think it cant be right - even it comes from steeeeveeee - still hate to say but could it be that this patch is a little buggy ?
------=_Part_228420_103341265.1372414817211
Content-Type: text/html; charset=utf-8
Content-Transfer-Encoding: quoted-printable
<html><head><style type=3D'text/css'>p { margin: 0; }</style></head><body><=
div style=3D'font-family: arial,helvetica,sans-serif; font-size: 10pt; colo=
r: #000000'>Hello everyone,<br><div style=3D"color:#000;font-weight:n=
ormal;font-style:normal;text-decoration:none;font-family:Helvetica,Arial,sa=
ns-serif;font-size:12pt;"><div style=3D"font-family: arial,helvetica,sans-s=
erif; font-size: 10pt; color: #000000"><div style=3D"color:#000;font-weight=
:normal;font-style:normal;text-decoration:none;font-family:Helvetica,Arial,=
sans-serif;font-size:12pt;"><div style=3D"font-family: arial,helvetica,sans=
-serif; font-size: 10pt; color: #000000"><br>After doing some testing in my=
Production system :-( i found something intresting.<br><br>I Use now OSB a=
gain, TOE, and the other standards together with Mysql, imple,mented by ama=
visd - Zimbra setup<br><br>So now if i use latest git Version i get Results=
like:<br><br><pre>X-DSPAM-Result: Spam
X-DSPAM-Processed: Fri Jun 28 04:14:52 2013
X-DSPAM-Confidence: 0.9076
X-DSPAM-Probability: 1.0000
X-DSPAM-Signature: 51ccf19c82621724820715<br><br>Which seems odd, i send th=
e same test message from gmail over and over again.<br>reclassify it retrai=
n it - training give PASS, still next time same thing again<br>X-DSPAM-Resu=
lt: Says spam, probability always 1.0000 <br><br>Then i testet 3.9.0 - beca=
use iam using this on zimbra 3.9.0 is the default installation (i jsut syml=
inked the git version so i can switch quickly)<br><br>The difference was ma=
ssive out of the box<br>without retrain the last message i got<br><br>X-DSP=
AM-Result: Innocent
X-DSPAM-Processed: Fri Jun 28 04:30:03 2013
X-DSPAM-Confidence: 1.0000
X-DSPAM-Probability: 0.0023
X-DSPAM-Signature: 51ccf52b148228935421196<br><br><br>Same Message same Sen=
der just about 30 seconds later after switched dspam Version. <br>only diff=
erence was dspam 3.9.0 instead of git.<br><br>Well i dont say of course thi=
s is a better result, i just say its looks more like what ife expected aft=
er retrain it 20 times as innocent.<br><br>Of course i checked if iam using=
the right database (dspam stats and so show me the right stats, because i =
switched to mysql theres no hashfiel anywhere so i think both version shoul=
d really work with the same database)<br><br>also as a side note, there was=
almost no difference with feature tb - off and tb=3D9 in the git version<b=
r><br>btw why is it marked as spam anyway with a confidence of 0.9 it shoul=
d be <br>my guess its because of probability is always 1.00000 , maybe some=
thing off there<br><br>So after some Digging in the History i see something=
about changed calculation in OSB and markov weighting - well exactly waht =
iam using<br>and yes there is a changed calculation method - <br>i just thi=
nk it cant be right - even it comes from steeeeveeee - still hate to say bu=
t could it be that this patch is a little buggy ?<br><br></pre></div></div>=
</div></div></div></body></html>
------=_Part_228420_103341265.1372414817211--
--===============4912529444302958014==
Content-Type: text/plain; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
------------------------------------------------------------------------------
This SF.net email is sponsored by Windows:
Build for Windows Store.
http://p.sf.net/sfu/windows-dev2dev
--===============4912529444302958014==
Content-Type: text/plain; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
_______________________________________________
Dspam-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/dspam-devel
--===============4912529444302958014==--