This repository includes cavity-based matrix factorization (CBMF) and approximate cavity-based matrix factorization (ACBMF) for matrix completion.

• Installing Almadillo, which is c++ library for linear algebra, is needed to run.
• Compiling and running was tested under gcc version 4.2.1 (macOS High Sierra).

Simply use make.

• -l: Learning rate. [default: 0.3]
• -L: Reguralization parameter. [default: 3]
• -R: Rank. [default: 10]
• -m: Maximum number of iterations. [default: 100]
• -r: Filename of dataset for training. [default: dataset/ml_1m_train.txt]
• -t: Filename of dataset for test. [default: dataset/ml_1m_test.txt]
• -o: Where output files are to be placed. [default: output/]
• -c: Exponent of convergence condition. If RMSE < pow(10, convint) is satisfied, it is regarded as convergence. [default: -5]
• -p: If this option is set, ACBMF is to be performed. Without this option, CBMF is to be done.
• -h: Show help.

Perfoming CBMF using 'dataset/ml_1m_train.txt' as training dataset and 'dataset/ml_1m_test.txt' as test dataset.

./cbmf -r dataset/ml_1m_train.txt -t dataset/ml_1m_test.txt

Perfoming ACBMF using 'dataset/ml_1m_train.txt' as training dataset and 'dataset/ml_1m_test.txt' as test dataset.

./cbmf -p -r dataset/ml_1m_train.txt -t dataset/ml_1m_test.txt

Showing help.

./cbmf -h