July 22, 2009

Well, we seem to have made a lot of progress since yesterday, or so Taylor tells me. Yesterday I figured out how to clean the torque data. It was really noisy, so first I cleared out any outliers, then I ran a running average smooth over it and the data actually looks really nice. I also took the derivative of the data. I started one of our computers in our cluster on cleaning all the torque data; it should finish sometime tomorrow afternoon. We have yet to figure out how to best crop our data (there is extra information before and after each interaction that is effectively 'silence'). Then we have to train our self-organizing map on the data, which Jivko got us code to do that yesterday, so it shouldn't be too hard. After that we're going to run a distance metric to compare all the data points. Matt (one of the grad students in our lab) spent last night writing us some code to do that pretty easily. All that's left to do after that is cross-validation with K-nearest neighbors and we have our results! I see the end, and it is near. Maybe there is a light at the end of the tunnel after all.