On 01/20/2014 11:51 AM, Matus UHLAR - fantomas wrote:
On 01/20/2014 11:37 AM, Matus UHLAR - fantomas wrote:
I heave received a ham message, that hit BAYES_50 at the time I received
it.
After feeding to "spamassassin -k" and rechecking it gives BAYES_99:
Unfortunately, feeding message to sa-learn explains it was not learntL
% sa-learn --ham < hamlearn
Learned tokens from 0 message(s) (1 message(s) examined)
Why has that happen? The message misses Message-Id, and it's
base64-encoded
text/html, but neither of those should cause the message not to learn,
should it?
I believe my BAYES db is fed up properly ;-)
On 20.01.14 11:42, Axb wrote:
tried --forget before feeding as --ham and compare results?
OK, an update: --forget and --ham do show that tokens were forgotten /
learned, but no change in result (score: 1.000)...
this probably means that the msg hits too many spammy tokens so that
"learning one" is not making the big difference.
Are you doing regular expiration runs?
