/**
* Compute the singular values of a matrix.
*
* @param A FloatMatrix of dimension m * n
* @return A min(m, n) vector of singular values.
*/
public static FloatMatrix SVDValues(FloatMatrix A) {
int m = A.rows;
int n = A.columns;
FloatMatrix S = new FloatMatrix(min(m, n));
int info =
NativeBlas.sgesvd('N', 'N', m, n, A.dup().data, 0, m, S.data, 0, null, 0, 1, null, 0, 1);
if (info > 0) {
throw new LapackConvergenceException(
"GESVD", info + " superdiagonals of an intermediate bidiagonal form failed to converge.");
}
return S;
}
/**
* Compute a singular-value decomposition of A (sparse variant). Sparse means that the matrices U
* and V are not square but only have as many columns (or rows) as necessary.
*
* @param A
* @return A FloatMatrix[3] array of U, S, V such that A = U * diag(S) * V'
*/
public static FloatMatrix[] sparseSVD(FloatMatrix A) {
int m = A.rows;
int n = A.columns;
FloatMatrix U = new FloatMatrix(m, min(m, n));
FloatMatrix S = new FloatMatrix(min(m, n));
FloatMatrix V = new FloatMatrix(min(m, n), n);
int info =
NativeBlas.sgesvd(
'S', 'S', m, n, A.dup().data, 0, m, S.data, 0, U.data, 0, m, V.data, 0, min(m, n));
if (info > 0) {
throw new LapackConvergenceException(
"GESVD", info + " superdiagonals of an intermediate bidiagonal form failed to converge.");
}
return new FloatMatrix[] {U, S, V.transpose()};
}
