Dark Days with Bogofilter

It took a month, but I finally had bogofilter trained up to some passable standard (I had no massive corpus of spam lying around). It was frustrating that it took so long seeming to only respond to the negative “this is spam, this is bad”, versus the positive “this is good email”. This runs counter to my understanding of how to train a Bayesian network normally, where you would want at least as many examples of the good email as of spam. I came to associate this feature of bogofilter’s in my mind with its violent, weird creator, and it drifted some being a cute toy, to somthing darker. That was probably unreasonable. ### Beginning of the End

But finally it was working. And then oneday it stopped. I don’t know if the database corrupted as I originally thought (but could find no reference to) or the nature of spam radically changed overnight or what. ### The Old Ways

So I gave up, I’m using Spamassassin now, and its worked beautifully out of box. 137 messages caught in the last 24 hours, 6 missed. Mark one up for diversity in tactics.