[Spambayes] how spambayes handles image-only spams

Meyer, Tony T.A.Meyer at massey.ac.nz
Wed Sep 10 12:36:20 EDT 2003


> >>> Bill Yerazunis wrote
> > How well does SpamBayes do on the SpamAssassin test corpus?
> 
> That's the stuff available here: 
> http://spamassassin.org/publiccorpus/ ?
> 
> I'm not aware that anyone's run against it - perhaps someone's got
some CPU 
> cycles burning a hole in their pocket and would like to try?

I occasionally run against this.  An example (with 'fresh from cvs'
spambayes):

filename:  no_sean_sas
ham:spam:  7580:7580
fp total:       44
fp %:         0.58
fn total:       16
fn %:         0.21
unsure t:      356
unsure %:     2.35
real cost: $527.20
best cost: $592.40
h mean:       3.40
h sdev:      14.19
s mean:      97.94
s sdev:       9.43
mean diff:   94.54
k:            4.00

(This was done a few weeks back, but I don't think that anything has
changed that would effect it since).  Note that I haven't gone through
the corpora to check that there isn't any misclassified mail.

=Tony Meyer



More information about the Spambayes mailing list