tuning [was: ignore text/plain ...]
David Relson
relson at osagesoftware.com
Wed Aug 13 02:22:24 CEST 2003
At 07:43 PM 8/12/03, David Flanagan wrote:
># int cnt prob spamicity histogram
># 0.00 100 0.009030 0.006626 ################################################
># 0.10 0 0.000000 0.006626
># 0.20 0 0.000000 0.006626
># 0.30 0 0.000000 0.006626
># 0.40 0 0.000000 0.006626
># 0.50 0 0.000000 0.006626
># 0.60 0 0.000000 0.006626
># 0.70 0 0.000000 0.006626
># 0.80 0 0.000000 0.006626
># 0.90 38 0.983336 0.444546 ###################
David,
FYI, using "-vv" with bogofilter directs it to output a histogram of the
tokens in the message. Above it the histogram for your message to the
list. As you can see, with my config file and wordlist, there were 100 low
scoring (hammish) tokens and 38 high scoring (spammish) tokens in the messages.
Below are the last two lines of your "-vvv" output:
N_P_Q_S_s_x_md 312 0.00e+00 3.08e-03 5.02e-01
1.00e-02 4.15e-01 0.100
The 0.100 is the min_dev value and the 4.15e-01 is the robx value (the
score for unknowns). These values tell me that you're using the default
configuration. Have you considered running the tuning/bogotune
script? Its purpose is to determine the optimal scoring parameters for
bogofilter, based on the messages in _your_ environment. I've been very
please with the script's results.
David
More information about the Bogofilter
mailing list