3/27/2008 9:02:00 PM
Another installment of the Almost Live series and the Bayesian Filter is completed. If you haven't been following along, one of the program requirements was to be able to filter past performance data automatically. The application, written using C#, uses a Bayesian filter to automatically filter the data and separate the data into "good" data and "bad" data, with later algorithms using the "good" data. It is a relatively simple Bayesian filter, but it demonstrates how to go about training a filter, the roll the filter's threshold can play, and discusses some areas to build a more complex filter. At the end of the day, it is has turned out to be a very powerful solution for filtering data, so if you are ever faced with a situation where you know what is good and bad and want to teach a computer how to approximate that same type of knowledge, a Bayesian filter could be the way to.
Source code available for download here. |
Double-click for full screen |
Subscribe