-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

John Gardiner Myers writes:
> DATE_IN_FUTURE_48_96 has an S/O of 1.0, yet is disabled.  Others have 
> S/O values that are just barely under the current threshold.
> 
> The results are nonsensical.  That highly suggests there is something 
> wrong with the criteria.

I wouldn't say it's nonsensical -- it's what we agreed on.

The criteria include a minimum number of hits in the corpus; it's
currently set to require 0.1% of spam hit.  If you look at the score-map
data:

http://buildbot.spamassassin.org/ruleqa?daterev=20060106-r358194-n&rule=DATE_IN_FUTURE_48_96&s_detail=1

  scoremap spam:  3  26.67%    8 **********
  scoremap spam:  4  16.67%    5 ******
  scoremap spam:  5  10.00%    3 ****
  scoremap spam:  6  10.00%    3 ****
  scoremap spam:  7   6.67%    2 **
  scoremap spam:  8   6.67%    2 **
  scoremap spam: 11   3.33%    1 *
  scoremap spam: 18  10.00%    3 ****
  scoremap spam: 19   3.33%    1 *
  scoremap spam: 20   3.33%    1 *
  scoremap spam: 26   3.33%    1 *

the 3rd column is the number of messages hit.  in other words, in a
138165-spam corpus, it's hit about 20 mails in total.

At some point, you have to say "this rule doesn't hit enough mail to
be worthwhile"; we've chosen 0.1% for that cut-off.  This rule is
hitting 0.02% of spam.

anyway, I've just checked in a change that'll allow hit-rates
all the way down to 0.02%.  why not. ;)

- --j.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)
Comment: Exmh CVS

iD8DBQFDvt7jMJF5cimLx9ARAkGyAJ9YlwmxfE2VU2b4lw+3d5/mR8/DtACgnVTn
LjYOnQBaKJLF1xPp/HaH9kI=
=3/ll
-----END PGP SIGNATURE-----

Reply via email to