On Tue, 30 Sep 2014, terrygalant.li...@fastest.cc wrote:
On Tue, Sep 30, 2014, at 10:57 AM, John Hardin wrote:
Did you run the above as user zimbra?
This time it is,
su - zimbra
$ /opt/zimbra/libexec/sa-learn --dbpath /opt/zimbra/data/amavisd/.spamassassin
--dump magic
0.000 0 3 0 non-token data: bayes db version
0.000 0 1150 0 non-token data: nspam
0.000 0 19454 0 non-token data: nham
0.000 0 120765 0 non-token data: ntokens
You might want to verify the "nspam" and "nham" numbers change as you feed
in messages for training.
Hm. I resubmitted a bunch that'd been trained, and saw no change
You wouldn't for messages already learned, unless you changed their
classification (e.g. was initially learned as ham, relearned as spam,
would relocate token counts from nham to nspam).
Doing it manually, outside of zimbra, I exported to text, then re-submitted 15
messages
By "re-submitted" you mean messages that were in the Zimbra training
corpora folder?
/opt/zimbra/libexec/sa-learn --dbpath /opt/zimbra/data/amavisd/.spamassassin
--dump magic
/opt/zimbra/libexec/sa-learn --dbpath /opt/zimbra/data/amavisd/.spamassassin
--no-sync --spam /opt/zimbra/TEMP/spam/*
Learned tokens from 15 message(s) (15 message(s) examined)
"Learned tokens" with a non-zero number suggests they *haven't* been
learned before. If you saw "learned tokens from 0 messages (15 examined)"
that'd indicate bayes had already seen the messages.
/opt/zimbra/libexec/sa-learn --dbpath /opt/zimbra/data/amavisd/.spamassassin
--sync
bayes: synced databases from journal in 1 seconds: 2189 unique entries
(2189 total entries)
Now appears to have changed
/opt/zimbra/libexec/sa-learn --dbpath /opt/zimbra/data/amavisd/.spamassassin
--dump magic
0.000 0 3 0 non-token data: bayes db version
0.000 0 1165 0 non-token data: nspam
0.000 0 19454 0 non-token data: nham
That suggests Zimbra *isn't* actually learning the messages being
submitted to the training mailbox.
Try submitting some new ham to the training mailbox and see if the numbers
change. (Note: I don't know *when* Zimbra does that learning, or whether
it can be initiated manually if the learning is on a schedule.)
Are there any logs indicating that Zimbra has attempted to learn
messages, any indicating it succeeded?
--
John Hardin KA7OHZ http://www.impsec.org/~jhardin/
jhar...@impsec.org FALaholic #11174 pgpk -a jhar...@impsec.org
key: 0xB8732E79 -- 2D8C 34F4 6411 F507 136C AF76 D822 E6E6 B873 2E79
-----------------------------------------------------------------------
All I could think about was this bear is so close to me I can
see its teeth. I could have kissed it. I wished I had a gun.
-- Alyson Jones-Robinson
-----------------------------------------------------------------------
4 days until the 10th anniversary of SpaceshipOne winning the X-prize
