Bayes filter not learning

[bberny]

My gateway runs for around a year now, but bayesian filter is still not learning spam well.

I have set threshold to

bayes_auto_learn_threshold_spam         7.0

in /usr/share/spamassassin/10_default_prefs.cf and also in /etc/mail/spamassassin but it still did not learn messages above 14 score.
Hams are correctly learned based on sa-learn --dump magic (7nspam and 395nham) and sometimes I can see it in logs that autolearn=ham.
But when I receive mail that is above my spam threshold (even 15 score of mail does not seem enough) it always shows autolearn=no.
Example below shows not learned mail that should be learned by bayes autolearn but is not

Dec 31 13:55:43 mailgw pmg-smtp-filter[27227]: 8140D61CEFDCEB39E6: SA score=9/5 time=0.513 bayes=undefined autolearn=no autolearn_force=no hits=BODY_PLASTIC_MOLD(1.9),DKIM_ADSP_CUSTOM_MED(0.001),DRUGS_ERECTILE(2.221),FORGED_GMAIL_RCVD(1),FREEMAIL_FROM(0.001),KAM_DMARC_STATUS(0.01),NML_ADSP_CUSTOM_MED(1.2),SPF_SOFTFAIL(0.972),SPOOFED_FREEMAIL(1.269),SPOOF_GMAIL_MID(1.054),T_SPF_HELO_TEMPERROR(0.01)

How can I get my autolearn to start learning based on my configurations ?

 

[BJ78945]

Hi,

autolearn works out of the box. But the score you show in your log is not the only thing what is necessary. Default the bayes_auto_learn_threshold_spam is 12. That means, mails with score >12 will be learned as spam. But the score must be generated from at least >3 points from header and >3 points from body. See https://spamassassin.apache.org/full/3.3.x/doc/Mail_SpamAssassin_Plugin_AutoLearnThreshold.html