Collective Matrix Factorization

This is a python implementation of Collective Matrix Factorization using Newton's method.

Input Data Format

In this code, each relation is stored as a matrix in coordinate format. There are some examples in data/.

row,col,value
394,264,2
464,201,3
111,198,4

Quick Usage

$ python3 cmf.py --train data/ml-1m/train.txt --test data/ml-1m/test.txt --user data/ml-1m/user.txt --item data/ml-1m/item.txt --out ml-1m.txt --alphas '0.5-0.5-0.5' --link log_dense --k 8 --reg 0.1 --lr 0.1 --iter 50 --tol 0

You have to use python3 to run this code. Type python3 cmf.py --help for more details about the parameters.

Reference

* Singh, Ajit P., and Geoffrey J. Gordon. Relational learning via collective matrix factorization. Proceedings of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 2008.