I started working on this about a month and a half ago. My RMSE is pathetic thus far(just started tweaking a kNN with Pearson correlation as a distance metric), but I don't really care. It's a great way to brush up on a lot of CompSci concepts at once you may be rusty on. Besides obviously machine learning, when you're trying to process 2 gb of training files on your personal computer, O(n) time/space complexity REALLY starts to matter, as does choice of implementation language...using mostly Python with C extensions for the heavy math right now...