Observations from my new training script

[Boris 'pi' Piwinger]

Hi!

So I have been away for ten days. Here is what happened in
the meantime.

I have not had a single false classification since Thursday
or so. I had up to one in the days before.

There is one major risk: If you correct a false negative,
this might trigger false positives, due to the small
database. It happened to me, but I have extremly daring
cutoffs. So I learned, to retrain with the complete database
after corrections. I did that now, which just added another
12 mails to my database, after just those very few
corrections from last week.

So it works for me *much* better than any other
configuration or training I had before.

pi