NMF-CR is a model to implement link prediction based on non-negative matrix factorization incorporate row space information and column space information.

(1) NMF_CR_general.py is a general version of NMF_CR, its loss function is defined as follows:

when $\alpha_{0}=1$, $\alpha_{1}=1$, $\beta_{0}=1$ , NMF_CR_general.py degenerates into NMF_CR.py

(2) utils.py contains some common used functions, e.g., train_test_split to split dataset into training data and probe data.

(3) Metrics.py contains common used metrics.

Take the dataset ASongIceFire as an example,

CR_train_ASongIceFire_*.py implement NMF_CR(combined row space information and column space information) on dataset ASongIceFire as low dimensionality $k$ varies.

R_train_ASongIceFire_*.py implement NMF_R( row space information ) on dataset ASongIceFire as low dimensionality $k$ varies.

C_train_ASongIceFire_*.py implement NMF_C( column space information ) on dataset ASongIceFire as low dimensionality $k$ varies.

train_ASongIceFire_*.py implement original NMF on dataset ASongIceFire as low dimensionality $k$ varies.

Folder output contain the results of all models.

For example, python CR_train_ASongIceFire_k256.py