How to check that FANN module is working?

[Marc Stürmer]

Hi,

I am struggling to eliminate some kind of resilient spam. Rspamd is 1.4.0 self compiled from Github with FANN enabled.

How can I check that the module is working as intended and what it is actually doing through the day?

tia

[Vsevolod Stakhov]

It should now write how many learns are left to start
meta-classification. But I'm curious why do you think that this module
will help you to resolve your issues. Maybe, I can provide you more
useful feedback if you explain what you are trying to achieve a bit more
detailed.

--
Vsevolod Stakhov

[Marc Stürmer]

Hi Vsevolod, 

thanks!

Well, what I am trying to achieve is to make an even sharper razor out of Rspamd so that I can get an even better spam detection rate. As far as I do understand that's the reasoning behind the neural network module, detecting patterns in the log file and associating it with spam and ham, therefore building the base for another metric value taken into account on the spam score calculation. Correct? 

My problem I try to solve is that one of my users recently opened one email containing sex spam, and now just gets flooded by such emails day by day, Spamassassin just didn't cut it anymore, that's why I switched over to Rspamd, the detection rate is much better but still 7-10 mails slip through the whole filtering setup. keeping my user constantly annoyed. 

The problem on those spam mails is that 

a) they seem to be from a botnet which hasn't been quite well listed in the RBLs yet and

b) are specifically designed to give very few triggers on spam detection per se, 

and she's the only user getting this specifig spam, so it's targeted at her. 

If you would like to help I could offer some samples on this kind of SPAM. 

[Vsevolod Stakhov]

I have strong doubts that ANN would help you to resolve that issue: they
usually require very intensive training before being useful. You can
contact me privately and send some spam samples so I could suggest you
something.

--
Vsevolod Stakhov