Bump! So is no one out there using fractional scoring?
There are several instances in my spam scoring that
fractional exponents would be particularly useful.
For example you want to discriminate against "Fwd:,
Fw:, etc.) but you would like to cap the maximum
weight for that alone. I.E. you don't want a message
filtered simply for that in the absence of any other
spam signs.
Daryle A. Tilroe wrote:
J. Hardin identified a problem with fractional base issues
a while ago in some of his recipes. I guess I have
confirmed that it is still an issue and, further, there
is a major problem with fractional exponent scoring as
well. Some of my observations:
John's 'good' value of .11 for a base doesn't work on my
version of procmail/redhat. It actually gives bad results
and .1 gives the correct ones. E.G. .1^1 with 100 matches
gives 10 but .11^1 with 100 matches gives <1. Actually
quite a few of the fractional bases in the range 0.01-0.99 are
flaky but there seems to be no pattern; for example .03
is fine and so are .66 and .05, but .55 is screwed. I find
this rather disconcerting from a computer :-/.
Fractional exponents are also completely buggered and do not
behave as procmailsc says they should, at least with bases
less than 3 or so. E.G. with 10 matches 1^.9 should be ~6.5
(asymptotically approaching 10) but procmail returns ~1 and
only goes up to one with large matches. What's the deal here?
It seems if you stray from 1^1 or non-integer bases and
exponents you better test it first to make sure things
work; because they often are broken. This scoring is
trivial and I would have hoped it was working as advertised
in a program as old a procmail. Quite disappointing :-(.
--
Daryle A. Tilroe
_______________________________________________
procmail mailing list
procmail(_at_)lists(_dot_)RWTH-Aachen(_dot_)DE
http://MailMan.RWTH-Aachen.DE/mailman/listinfo/procmail