Re: train-good and train-spam; cron error

just chipping in for the record,

on my server i still have some accounts that uses BDB (~30) and some
that uses PBL. once in a while one process gets stuck (im assuming on
the lock) that causes a buildup of processes for that account and
eventually spamprobe doesn't work for ANY account (including the PBL
ones)

i can't explain it, but that's just a gotcha to be aware of if you have
multiple accounts with BDB databases on the same server (i've never seen
that problem with PBL, and there's a lot more of them)

note that i'm NOT using a shared database (-D) and i'm still using
spamprobe 0.9b (for BDB) and 0.9e (for PBL) 

On Fri, 2004-12-10 at 13:17, Brian Burton wrote:
> John Chandler wrote:
> >     The documentation still recommends using "good" and "spam" 
> >     for the initial training and for later corrrective training.
> 
> If you use train for incoming email then you should use train-good and 
> train-spam for corrections.  Using the good/spam commands would not hurt 
> anything but could increase database size.
> 
> 
> >     Would the use of "train-good" or "train-spam"
> >     in initially training spamprobe 
> >     or in later corrective sessions be appropriate,
> >     in order better to control the size of the database?
> 
> Once 1.1 is finished (needs documentation changes prior to release) the 
> proper way to build an initial database will be the auto-train command. 
>   Until then I recomment using the good and spam commands.
> 
> auto-train will do its best to minimize the database size during 
> training and will train on your saved hams and spams in a balanced way 
> that should improve starting accuracy.
> 
> 
> >     If not, could "train-good" and "train-spam" perhaps be used 
> >     for just the latter part of the initial training?
> >     (That is, train on one small "good" file and one small "spam" file,
> >     then switch to "train-good" and "train-spam" and train some more.)
> 
> As long as you mix up your mail using train-good and train-spam would be 
> fine.  By that I mean if you just have two buge files (mail.good and 
> mail.spam) then using train-good and train-spam will give poor accuracy 
> to start but if you have 5 or 6 ham files and 5 or 6 spam files then 
> training with train-good and train-spam should work well if you 
> alternate them (i.e. train one on good file, then one spam file, repeat)
> 
> 
> > 2)  My cron job ran fine for seven months,
> >     but recently it has given the following error in every run.
> 
> Your sp_words file is corrupt.  If you retrain and see the exact same 
> error then maybe you have an email that is crashing SP?  If that's the 
> case and it's reproducible then let me know and I can try to fix the bug.
> 
> 
> >     We are running spamprobe 1.0 on Solaris 9.
> >     Any suggestions of things to try?
> 
> Build a new sp_words file and run db_verify on it to see if it's 
> corrupted immediately after the rebuild.  If it is then we'd want to 
> know what's causing that to happen.  Note that BDB is vulnerable to 
> problems if you ^C during an update.
> 
> You might want to give the latest 1.1x release a try.  It's very stable 
> and has 1.0's features but a better (I think) parser than 1.0.  It also 
> has some improvements that try to clean up the database more cleanly in 
> the event of improper termination of SP during an update.
> 
> If possible switching to PBL would be a good idea.  PBL seems to be much 
> more robust than berkeley db.  I'm thinking of adding QDBM support over 
> the holiday to see if it's a reasonable alternative to PBL and BDB.
> 
> All the best,
> ++Brian
> 
> 
> -------------------------------------------------------
> SF email is sponsored by - The IT Product Guide
> Read honest & candid reviews on hundreds of IT Products from real users.
> Discover which products truly live up to the hype. Start reading now. 
> http://productguide.itmanagersjournal.com/
> _______________________________________________
> Spamprobe-users mailing list
> Spamprobe-users-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org
> https://lists.sourceforge.net/lists/listinfo/spamprobe-users



-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now. 
http://productguide.itmanagersjournal.com/