Steve wrote: > Hallo Tom, > > >> Based on an ebuild I found in the dspam bugtracker, I created two >> ebuilds (3.9.0_alpha2 and a git version) which are available from my >> private overlay. They both work for me, and 3.9.0_alpha2 is running in >> production environment on my mail server. You're free to use them at >> your own risk (btw: I am not a gentoo developer): >> > Tell me more. What is your experience with DSPAM 3.9.0? Anything good? > Anything bad? How many mails do you process with DSPAM 3.9.0? How is the > speed? How is the accuracy? And, and, and... I would really love to hear some > feedback.
The machine runs a private mail server, so there is not that much mail
to process. A few weeks ago I switched the backend from mysql to sqlite,
so I started training from scratch using teft + osb.
Current results:
TP True Positives: 99
TN True Negatives: 924
FP False Positives: 2
FN False Negatives: 61
SC Spam Corpusfed: 0
NC Nonspam Corpusfed: 0
TL Training Left: 1574
SHR Spam Hit Rate 61.88%
HSR Ham Strike Rate: 0.22%
PPV Positive predictive value: 98.02%
OCA Overall Accuracy: 94.20%
The speed is actually not that great, but probably due to the setup: no
server (client-only), and the use of sqlite.
Precision rates are a lower than with the previous setup (dspam
3.8.0/mysql/teft+word) but that's likely because the training data needs
to mature a bit more.
It runs fine currently, I see no dramatic changes between 3.8.0 and 3.9.0.
>
>
>> Available through: svn checkout
>> https://svn.whyscream.net/whyscream-overlay/testing/mail-filter/dspam/
>>
> Great but I know Marko :) he will not use it as long as it is not in Portage.
That's fine :) It might be useful for other Gentoo users, though.
>
>
>> --
>> Regards,
>> Tom
>>
> Steve
--
Regards,
Tom
signature.asc
Description: OpenPGP digital signature
------------------------------------------------------------------------------
_______________________________________________ Dspam-user mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/dspam-user
