Daniel Alabi
Cody Wang
CS 324
Our project was to implement and analyze regularized SVD in python, as suggested by Simon Funk and others [1,2].
For a more detailed explanation of the regularized SVD technique we use and also some analysis, see our paper, finalproject.pdf.
We have 5 main python files:
-
regularizedSVD.py - contains implementation of SvdMatrix that reads in ratings and trains the UV matrices based on these ratings. Reports the final train RMSE. Can also read in test ratings and report the final test RMSE.
-
plotsSVD.py - used to plot graphs used to pick the best parameters.
-
writeplots.py - used to write results of varying different parameters used in our SvdMatrix class to a file.
-
userbased.py - contains implementation of user-based collaborative filtering technique. This is used as a control. We compare how good/bad our technique is against this.
-
itembased.py - contains implementation of item-based collaborative filtering technique. Also used as a control.
-
scipySVDcontrol.py - contains implementation of SVD technique using scipy SVD, filling in the partial utility matrices with averages.
See the contents of the above files for more information on how to run them.
[1] http://www.timelydevelopment.com/demos/NetflixPrize.aspx