Projects

• Get advance warning of U.S. recessions 6 to 12 months before they hit at recessionprediction.com.

• The Naive Bayes classifier performs remarkably well on many classification data sets, especially considering the patently unrealistic assumptions it makes about the data. In my college thesis, I investigated the reasons why Naive Bayes does so well and how we can take advantage of these ideas to build better classifiers.

  You can read the full paper here (68 pages) or see a poster summarizing the work here.

Future Projects (?)

• There are only a finite number of words. Therefore, if I run a word through Google translate enough times, I have to eventually hit a cycle -- the system can't be chaotic.
  
  That said, does the speed with which I reach a cycle say anything about the word? Are the cycles short or long? If they're short, is it because the translation fidelity is high, or is it because a few words act as attractors for a lost majority of words? Can I study language this way, or am I only studying Google translate?
• If I treat the thesaurus as a digraph, what words are most connected? Which word should I pick for a "six degrees of Kevin Bacon" type game?
• I'd like to put together a simple Django app for managing ultimate frisbee pickup games, possibly built on top of the Twilio API. Getting a dozen people to reliably show up at the same place every week can't be that hard, right?
  
  I'd probably host on Heroku. This is 1-2 day project.
• I'd like to combine Bump with the Dropbox API to enable easy file sharing off a mobile device. Could be cool, but requires a skillset that I don't really have.
  
  Another cool thing you could do with the Dropbox API and phones: Take a picture of a card, get a file in your Dropbox. File sharing via card sharing.
• I'd like to write a poker bot. Maybe in Obj-C? Has anyone done that?