brede_mat_reconerror - Reconstruction error for a matrix E = brede_mat_reconerror(X, L, M, R) E = brede_mat_reconerror(X, L, R) Input: X Original matrix as a 'mat' structure L Left matrix as a 'mat' structure M Middle matrix as a 'mat' structure R Right matrix as a 'mat' structure Property: Subspaces [ {1:size(M.matrix,1)} | nonnegative integers ] The subspaces the reconstruction error is computed in. Output: RE Reconstruction error as a vector Computes a reconstruction error, that is the difference between an 'original' matrix X and the product of its decomposed matrices L*M*R (or L*R). U = X - L*M*R or U = X - L*R The reconstruction error is computed as the Frobenius norm, ie, the trace of the square error matrix trace(U'*U) RE = norm(U, 'fro')^2 The reconstruction error is by default computed for subspace from 1 to the size of the decomposed subspace and returned as a vector with the first element being the reconstruction error for a subspace with dimension one using the first column in L and first row in R. See also BREDE, BREDE_MAT, BREDE_MAT_NMF, BREDE_MAT_SVD. $Id: brede_mat_reconerror.m,v 1.4 2004/10/14 17:03:55 fnielsen Exp $

Generated on Fri 27-Nov-2009 18:11:22 by