Wednesday, January 12, 2011

Machine Learning Contests

I just found another machine learning contest site called TunedIT. TunedIT seems to be very similar to Kaggle. On first inspection one of the TunedIT contests seemed to be very interesting - categorizing the genre and instruments in music. My current area of research involves one dimensional time-series data, so I thought audio would be an interesting domain to work with. Unfortunately, instead of providing the raw audio data, the contest organizers decided to process the audio and provide feature vectors instead. No longer being a one-dimensional temporal problem, the contest has lost much of its appeal.

I have been treating the traffic prediction contest on Kaggle as a one-dimensional temporal prediction problem (considering every road segment as a completely independent problem). I am currently ranked 8th out of 214 teams. I am guessing that my ranking won't improve before the contest end because I don't have any new ideas on how to increase my score. Most of my submissions so far have been used for parameter tuning. However, my parameter tuning efforts seem to have reached a local optimum.

No comments: