Incremental Singular Value Decomposition-like Matrix Factorization for Collaborative Filtering using Stochastic Gradient Descent -------------------------------------------------------------------------------------------------------------------------------- Based on Simon Funk's Netflix implementation [1-4] which processes one latent feature at a time. Uses L2 regularization to control the magnitude of the user-feature (P) and item-feature (Q) matrices, based on [3]. General SVD background can be found at [5]. Optional learning rate decay, e.g. [8]. Compile ------- Includes the Eigen C++ headers [6]. Modify the include flag in the Makefile to use a different version. Requires Boost libraries (runtime and development headers). On Ubuntu, 'apt-get install libboost-all-dev', on OS X, 'brew install boost' Compile with 'make' Test with 'make test' Sample Output ------------- bin/svd_collabfilt -o data/predictions.tsv -i data/training.tsv Input file: data/training.tsv Output file: data/predictions.tsv Predictions: 25 Latent features: 240 Threads: 4 Read file data/training.tsv (282931 entries) Users with data: 33217 Items with data: 8319 Writing user-feature matrix data/P_demo.mat ... Writing item-feature matrix data/Q_demo.mat ... Writing predictions file data/predictions.tsv ... The predictions file contains the top 25 recommendations. First column is the user, then the item, then the rating. Links ----- [1] http://sifter.org/~simon/journal/20061211.html [2] http://www.timelydevelopment.com/demos/NetflixPrize.aspx [3] http://www.netflixprize.com/community/viewtopic.php?id=1423 [4] http://www.netflixprize.com/community/viewtopic.php?id=481 [5] http://www.quuxlabs.com/blog/2010/09/matrix-factorization-a-simple-tutorial-and-implementation-in-python/ [6] http://alias-i.com/lingpipe/docs/api/com/aliasi/matrix/SvdMatrix.html [7] http://eigen.tuxfamily.org/ [8] http://bengio.abracadoudou.com/lectures/old/tex_ann.pdf timnugent@gmail.com