Ben Jackson wrote: BJ> The problem with the GA is that it has everything to do with the spam BJ> corpus it runs on and nothing to do with any other spam. It solves the BJ> problem posed to it too literally for our needs. The intelligibility BJ> of the results is poor, which makes it hard to determine if they are BJ> right or wrong.
Well, there are a number of measures in there to try and make sure that it doesn't over fit to the corpus, but retains generality. There are better ways of doing it probably, but it's pretty good the way it is. The intelligibility of the result is substantially improved if you read through the freqs and analysis files I sent out yesterday or whennot. C _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
