Since there's been discussion on advogato diaries lately about which projects are worth donatiing to, I'd like to point out that I'm accepting donations for my work on BitTorrent, a project I've been working on full-time for over a year, with a mature, deployed piece of code to show for it.

Retroactive Spam

I realized that tiebreak behavior was unspecified in the spam filtering code I gave yesterday. The way I implemented it happened to work out that if there were 15 max probability spam and 15 max probability not spam, it assumed spam. A much more robust approach is, rather than summing up the fifteen scores whose absolute values are the highest, sum up the eight largest positive and eight most negative ones. I've now retroactively changed by previous diary entry to work that way. You can do that in a weblog.