Versions Compared


  • This line was added.
  • This line was removed.
  • Formatting was changed.
Comment: converted to 1.6 markup

Using 'overlap'

In the SpamAssassin "masses" directory, there's a tool called 'overlap', which is used to determine how much the rules in the ruleset overlap with each other.

For example, let's say I have a log file in spam.log, and want to examine how much the rules that start with _T_DRUG_ overlap with each other. I run overlap like so:

No Format
  ./overlap spam.log  > ov
  pcregrep "\sT_DRUG.*,T_DRUG" ov | sort -r +1 -n

Which in this case produces this output:

No Format
87      1.000   0.690   T_DRUGS_SLEEP_EREC,T_DRUGS_SLEEP
87      1.000   0.084   T_DRUGS_SLEEP_EREC,T_DRUGS_ERECTILE
328     1.000   0.715   T_DRUGS_ANXIETY_EREC,T_DRUGS_ANXIETY
315     1.000   0.887   T_DRUGS_DIET_EREC,T_DRUGS_DIET
315     1.000   0.304   T_DRUGS_DIET_EREC,T_DRUGS_ERECTILE
311     1.000   0.523   T_DRUGS_PAIN_EREC,T_DRUGS_PAIN
311     1.000   0.300   T_DRUGS_PAIN_EREC,T_DRUGS_ERECTILE
289     1.000   0.630   T_DRUGS_ANXIETY_OBFU,T_DRUGS_ANXIETY

Explanation of the columns: the first number is how many mails hit both rules; the second, how much of the hits for the first rule also hit the second; the third, how much of the hits for the second rule also hit the first.

So in the case of this line:

No Format
87      1.000   0.690   T_DRUGS_SLEEP_EREC,T_DRUGS_SLEEP

87 mails hit both rules; all of the mails that hit T_DRUGS_SLEEP_EREC also hit T_DRUGS_SLEEP; and 69% of the mails that hit T_DRUGS_SLEEP also hit T_DRUGS_SLEEP_EREC.

Overlap is very useful, if you believe that some rules are all hitting the same spam messages.