After finishing an intro to ML course, I felt inspired to implement a collaborative filtering algorithm right where it'd be most useful - in the database itself. As of now Sveddy implements only one algorithm: UV decomposition (similar to SVD, which inspired the project name). It also implements continuous learning to efficiently update a model as users express their preferences about items. Despite its simplicitly, it performs quite well on the Netflix Prize dataset, which I am proud of. However, there is still work to be done on the training algorithm's performance - it takes 6 minutes to train on 100 million ratings. I'd appreciate any insight on how to improve this.
Constructive criticism is welcome!