tuning.sh [was: bogofilter-0.13.6.3 - new current release]
David Relson
relson at osagesoftware.com
Fri Jun 20 14:03:08 CEST 2003
At 07:46 AM 6/20/03, Boris 'pi' Piwinger wrote:
>David Relson wrote:
>
> > 0.13.6.3 2003-06-18
>
> > * Fixed several errors in tuning/tuning.sh and reformatted
> > "Top 10 Results" output.
>
>So I did a first run:
>
>Top 10 Results:
> robs min_dev spam_cutoff run0 run1 run2 total
>0.0100 0.450 0.500000 40 44 40 124
>0.0320 0.450 0.500000 39 45 40 124
>0.0100 0.425 0.500000 47 52 44 143
>0.0320 0.425 0.500000 50 53 46 149
>0.1000 0.450 0.500000 54 55 53 162
>0.1000 0.425 0.500000 56 59 51 166
>0.0100 0.400 0.500000 60 60 53 173
>0.0320 0.400 0.500000 69 65 60 194
>0.3200 0.450 0.500000 68 64 63 195
>0.0100 0.375 0.500000 67 68 63 198
>
>Looks like I need to change the target. Is there a rule of
>thumb how to choose it? Can this be done automatically?
>
>Also I am surprised, it is set to 12 and returned runs with
>12 to 16. I thought it should be around 4?
>
>pi
pi,
There was a change to the printed detail lines. tuning.sh used to print
$fpos/3 (as an integer). As this loses information about fractional
values, I changed it so that it now prints the actual count (without the
division).
The rule of thumb is that $target should be 0.1% to 0.3% of the test set
size. When I've encountered problems with spam_cutoff being too uniform, I
change the value to see what happens. The bad news is that this is very
"seat of the pants", not scientific. The good news is that we now have
Greg's comprehensive bogotune script which covers much more ground than
tuning.sh does.
David
More information about the Bogofilter
mailing list