Code covered by the BSD License
- IHT_MC(y,M,sizeX,err,x_in...Matrix Completion via Iterated Hard Thresholding
- IST_MC(y,M,sizeX,err,x_in...Matrix Completion via Iterated Soft Thresholding
- NonCVX_MC(y,M,sizeX,p,err...Non-convex matrix Completion via Iterated Soft Thresholding
View all files
Matrix Completion via Thresholding
17 Jan 2010
Convex, Non-convex and NP hard matrix completion algorithms
Watch this File
Contains three matrix completion algorithms and a demo script for running them. Also compares against other matrix completion algorithms - Singular Value Thresholding and Fixed Point Iteration.
Solves the following three optimization problems:
min rank(X) subject to ||y - M(X)||_2<err via Iterated Hard Thresholding
min nuclear-norm(X) subject to ||y - M(X)||_2<err via Iterated Soft Thresholding
min ||S||_p subject to ||y - M(X)||_2<err, where S = svd(X) via Iterated Soft Thresholding
Requires Sparco since the masking operator has been defined in according to the Sparco framework.
The algorithms are general enough to work with any other linear operator, and not only the masking operator. The masking operator is just a special case when the problem boils down to one of matrix completion.
For comparing the results with other algorithms download the Singular Value Thresholding toolbox
MATLAB 7.8 (R2009a)