Steffen Müller's TYPO3 blog

OSBF-lua rules

Last week, I moved my IMAP account from university to a privately hosted server. At the same time, I said goodbye to Spamassassin and switched to OSBF-lua - the results were impressive...

Already installed by Michael, I just had to create some maildrop rules to feed OSBF-lua. After training the bayes database with 42 pieces of spam and 9 of ham, almost any following spam was classified correctly as bulk. The training was easy. Besides some perl/shell scripts, the bayes filter can be trained by sending false pos/negs back to OSBF-lua. You simply have to change the 'Subject:' and 'To:' headers (remember majordomo?). The first thing that came to my mind was: Hey, this could easily be done with a one-click button in Thunderbird. Any extension out there?

Statistics for nonspam.cfc:
-----------------------------------------------
Database version:                    OSBF-Bayes
Total buckets in database:                94321
Buckets used (%):                           9.0
Bucket size (bytes):                         12
Header size (bytes):                       4092
Number of chains:                          7444
Max chain len (buckets):                      5
Average chain length (buckets):             1.1
Max bucket displacement:                      3
Buckets unreachable:                          0
Trainings:                                    9
Classifications:                             88
Learned mistakes:                             7
Extra learnings:                              0
Ham accuracy (%):                         92.47
-----------------------------------------------

Statistics for spam.cfc:
-----------------------------------------------
Database version:                    OSBF-Bayes
Total buckets in database:                94321
Buckets used (%):                          62.6
Bucket size (bytes):                         12
Header size (bytes):                       4092
Number of chains:                         16430
Max chain len (buckets):                     60
Average chain length (buckets):             3.6
Max bucket displacement:                     29
Buckets unreachable:                          0
Trainings:                                   42
Classifications:                            134
Learned mistakes:                             2
Extra learnings:                              1
Spam accuracy (%):                        98.45
-----------------------------------------------
Spam rate (%):                            58.11
Global accuracy (%):                      95.95
-----------------------------------------------