Welcome
MCNF is a Matlab package for solving low-rank matrix completion problem with nonnegative factors. The problem can be described as follows. Assme that a matrix M of dimension m x n and with rank r can be (approximately) factorized as a product of two nonnegative rank-r matrices X and Y. Usually r << min { m, n }. Now suppose we observe partial entries of M. How can we recover the whole matrix M? MCNF applies the classic alternating direction method of multiplier to solve
min ||XY-Z||_F^2, subject to X >= 0, Y >= 0, and Z(i, j) = M(i, j) for (i, j) in Ω
Authors
Yangyang Xu, Wotao Yin, Zaiwen Wen, and Yin Zhang
Technical Report
Yangyang Xu, Wotao Yin, Zaiwen Wen, and Yin Zhang. An Alternating Direction Algorithm for Matrix Completion with Nonnegative Factors. Rice CAAM Technical Report TR 11-03. [pdf]
Download
MCNF (for matrix completion with nonnegative factors only) is distributed under the terms of the GNU General Public License. [Download]