Hi,

On Jun 13, 2015, at 5:37 PM, RW wrote:

> On Thu, 11 Jun 2015 19:06:50 -0400
> Phil Stracchino wrote:
>
>> On 06/11/15 16:07, Al Zick wrote:
>
>>> At first, it looked
>>> like dspam was working, but then I realized that I could not
>>> retrain. After doing more research I found where people were saying
>>> that you can not retrain with hash and toe, so I switched to tum.
>
> I don't know for sure, but it sounded like that was just a
> bootstrapping problem that could be solved by creating the database
> with training from corpus.

I have been creating the database with:
cat 782081.emlx | dspam --mode=tum --process --deliver=stdout --user  
antispam --client

This last time I created the database with:
cat dspam_training/spam/new/781097.emlx | dspam --user antispam -- 
class=spam --source=corpus --deliver=summary

How should the database be created?

> It's not a bad idea to do that anyway because TOE doesn't turn-on  
> until
> there are 2500 non-spams in the database.

With TOE it filters very little spam. With TUM it gets almost  
everything, however it usually crashes after 12 hours.


>> Why are you using hash storage?  Even if you don't have a real DB
>> engine installed that you can use, is sqlite not an option for some
>> reason?
>
> I frequently rebuild from corpus. Training ~9000 emails takes
> ~ 3 minutes with the hash driver and ~ 7 hours with sqlite3.

That is an incredible difference.

> I can't say I've ever noticed any corruption problem, and because  
> of the
> corpora I don't regard the css files as precious, or care about token
> expiry. Is there any particular reason why I shouldn't use the
> hash driver?

I moved dspam from procmail and put it into my postfix master.cf. I  
am hoping this will fix the css corruption problem.

Kind Regards,
Al



------------------------------------------------------------------------------
_______________________________________________
Dspam-user mailing list
Dspam-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/dspam-user

Reply via email to