bogus bogotuning

[Jason A. Smith]

On Wed, 2004-01-28 at 06:40, Greg Louis wrote:
> I argue, precisely, that anyone who understands what bogotune does and
> how it works would not use this option.  With that few messages, the
> user is safer to stick with bogofilter's default parameter values (or
> with values obtained elsewhere on a reasonably sized corpus).  Although
> anyone can make the one-line change that David suggested, I think it
> sends the wrong message altogether if we offer it officially.

The only objection I have seen anyone raise is that it will most likely
produce garbage results causing people to complain.  All I am asking is
that you allow the user to run it with a suitable warning, which will
prevent the complaints and still allow the user to experiment with the
command without patching and recompiling while they are waiting to
accumulate more email messages in their archive.

The only thing bogotune does is produce suggested parameters, right? 
That isn't a dangerous thing, especially if a warning comes with the
suggestions.  Why not allow it to run on any number of messages?  You
can even print a warning with a quality rating that indicates how
accurate the suggestions are based on how many messages the user has.

I have already had to tweak the default parameters in order to achieve
better results, but I am only able to guess at numbers since you won't
allow me to run bogotune to even just experiment a little.


On Wed, 2004-01-28 at 06:44, Greg Louis wrote: 
> Nor do you believe I am telling you that.  In fact, using bogotune on a
> corpus of 2,000 spam and 2,000 nonspam in the training db and 500 of
> each in the test set will give results that are far from perfect and
> should not necessarily be preferred to the defaults supplied with the
> distribution.  However, one has to start somewhere, and as the author
> of the original version of bogotune, I feel that's a reasonable
> minimum, and I don't think we should officially support going lower.

Your own statements admit that the minimum is quite arbitrary.  All I want
to do is to be allowed to experiment with bogotune a little while I am
collecting more email to get better statistics.

Why are you guys being so stubborn and so unwilling to listen to my point
of view?  What is so wrong with a strong warning instead of a fatal error?

~Jason