[Spambayes] No Defined Boundary

[Richard B Barger ABC APR]

Michael Kimball wrote:

Do you have Advanced Options | Header Options | Add probability (score) header set to 'Yes'?

RBB:  Thanks, Michael; good suggestion.  Yes, I display the probability to 6 decimal places.
Watching those was the reason I moved my spam probability and ham probability figures to such
aggressive numbers (0.39 and 0.01, respectively).  It appears that my ham mail stream is
well-trained and very uniform, because that extremely low setting rarely gives me any
misclassifications.

But I need to spend more time looking at those data.  I believe it was Tony Meyer who sent a
friendly warning that 1) I was entering dangerous territory (my words, not his) with those
settings, but 2) if it ain't broke, ...

Because of the volume of Unsures I'm still getting, I probably should revisit the spam
probability scores and see if I want to live even more dangerously, in order to reduce the
number of Unsures still further.  What's the good in having all this computing power if I just
use the box for a paperweight?  Let the computer compute!

Michael:  Now I spend 1 or 2 minutes checking that SpamBayes got it right, and 20 - 30 minutes
reading the SpamBayes list!!

RBB:  <g>

Thanks again.

Rich Barger
Kansas City

---

> Richard B Barger ABC APR wrote:

<snip>

> > I'd be delighted to be at 2 or 3 unsures; on a typical day, I get 1400 or 1500 messages,
> > 80+ percent spam or unsure, and most of the balance are postings to active newsgroups I
> > follow in real time.
> >
> > My settings are an aggressive 0.39 spam probability and 0.01 ham.  It is rare to get a ham
> > in the spam folder (I still visually scan it, even though it has some 1100 spam msgs a
> > day; it has processed 66670 msgs so far); I still get a few spams in my ham mail stream.
> > I haven't checked this accurately in a couple of weeks, but I'm probably still getting 6
> > or 7 percent unsures.
>
> Do you have Advanced Options | Header Options | Add probability (score) header set to 'Yes'?
> It might give you some clues about how to adjust your cutoffs, or which specific emails are
> the ones that need to be trained on.  In my case, not only do I get relatively few 'Unsure's
> (1 unsure in 26 total today: approx 4%?), but the score shows most of them just missed the
> cutoff, i.e. Ham scored as unsure has very low scores, close to the 0.2 cutoff, while Spam
> scored as Unsure has a score close to my 0.85 cutoff.  MOST Unsures that have scores close to
> .5 do have enough of both characteristics that they would probably always show up as Unsure,
> regardless of my cutoffs.
>
> Also most of my Spam shows scores close to 100%.  Mine may be a special case, as I get
> relatively little email, and most of it is spam.  Most of my Ham is from this list or a couple
> of others I'm on.
>
> I used to spend 10 - 15 minutes picking through my email and deleting what was spam manually.
> Now I spend 1 or 2 minutes checking that SpamBayes got it right, and 20 - 30 minutes reading
> the SpamBayes list!!
> %^)