new writeup re varying Robinson's s and the minimum deviation
David Relson
relson at osagesoftware.com
Sat Mar 29 18:59:47 CET 2003
Greg,
Great results! It's nice to see the flatness shown in the graphs. With
previous experiments using smaller corpora, the results have been highly
sensitive s and md values. A small change in either gives a big change in
effectiveness. As bogofilter's algorithm is generally quite effective in
production use, the sensitivity to smalll changes has bothered me.
Given your finding that mindev around 0.35 and s around 0.1 is in the
middle of the "good" range, I'm thinking of testing mindev =
0.25,0.30,0.35,0.40,0.45 against s = 0.32, 0.1, 0.032, 0.001.
Given that I have fewer bogomips available for the test, it will be some
hours before I have results. It _will_ be interesting to see if your
findings can be confirmed with other corpora.
Cheers!
David
More information about the Bogofilter
mailing list