relson at osagesoftware.com
Mon Jan 26 08:09:08 EST 2004
On Mon, 26 Jan 2004 13:57:26 +0100
Boris 'pi' Piwinger wrote:
> David Relson wrote:
> >> Well, it does not give the counts. But if this is considered
> >> enough, we could well delete the -p option in bogoutil in
> >> our cleanup.
> > -p is useful for finding a token's score. The fact that it uses the
> > default robs and robx values does, indeed, affect the score.
> So what is that score useful for if people don't use the
> default (which they should not).
It answers the question: How spammish (hammish) is this token?
Do you really care if the tokens score is 0.567890 or if it's 0.598654?
My interest lies in whether its down around 0.1 or whether it's up
around 0.99. The whole middle range is unimportant to me.
> > Remember
> > that the robs and robx parameters are used for biasing the score of
> > unknown (or infrequently trained) tokens to improve the overall
> > score of a message. Also as a token's counts increase, the effect
> > of robs and robx lessens.
> All correct, but why does that make a misleading valued useful?
It can be useful even though it's not 100% accurate.
I assert that most people use the default values of robs and robx, so
the value _is_ accurate.
More information about the Bogofilter