scikit-cuda
scikit-cuda copied to clipboard
eigsh
Is there any way to extract k-first eigenvalues (eigenvectors) rather than extracting the whole spectrum? This is the functionality eigsh provides in scipy. Currently skcuda.linalg.eig calculates all eigenvalues but this is a much harder problem then getting the first k, especially for high dimensional problems.
MAGMA provides a means of finding the eigenvalues of a symmetric/Hermitian matrix in an interval, but that functionality has not been integrated with skcuda.linalg.eig().
Ah, unfortunately don`t see cheevx being implemented in what looks like the only python wrapper pymagma @lebedov , any chance this would be added into skcuda future releases ?