Ran across this on hacker news earlier today, seems interesting and relevant.

http://martin.zinkevich.org/rules_of_ml/rules_of_ml.pdf