MatrixImpute

This is the Julia Repository for general low-rank Matrix/Tensor Completion (with/without Side Information). The main interface is:

Impute(A,k;method=:fastImpute,kwargs)

• A is the Matrix of size n x m/ Tensor of size n x m x p with missing entries. The entries missing should be of type Missing.
• k is the desired integer rank of fitting. For Tensors, this is currently assumed to be the canonical CP/Kruskal rank.
• kwargs are named optional arguments to specify hyperparameters for each method. For the specific names of the hyperparameters, please see the methods listed below.
• method is the keyword for specifying the low rank imputation method. Currently the following methods are implemented:
  • fastImpute (Fast Exact Matrix Completion: A Unifying Optimization Framework).
    • The hyperparameters are: lr (learning rate), B (side information matrix on the columns of size m x s, default none), approx (T/F, whether approximately low rank or perfectly low rank), and γ (regularization parameter).
  • fastImputeT (Tensor Completion based on fastImpute)
    • The hyperparameters are: lr (learning rate), and γ (regularization parameter).
  • (Not Completed) softImpute (Matrix Completion and Low-Rank SVD via Fast Alternating Least Squares).
    • The hyperparameter arguments are: λ (regularization parameter).

Current version is tested on Julia v1.2.

R package fastImpute would be available soon.