I haven't coded it up yet, but usually when it seems that spambayes'
performance is degraded, it means I've misclassified something. These
messages tend to stay misclassified when training to exhaustion, even
after 9 or 10 rounds. I'm contemplating an extension to tte.py that
will throw them back into my unsure folder and start over if training
doesn't stop on its own before N rounds due to everything being properly