Trouble with eigenvalues and lanczos-tag

[wolfgangp]

When I do:

import pyviennacl as cl
gpuA = cl.Matrix( np.diag((1.,2.,3.)) )
tag = cl.linalg.lanczos_tag(num_eig=3)
EigVals = cl.linalg.eig(gpuA, tag)

I get:

ArgumentError: Python argument types in
    pyviennacl._viennacl.eig(matrix_row_float, lanczos_tag)
did not match C++ signature:
    eig(viennacl::matrix<float, viennacl::column_major, 1u>, viennacl::linalg::power_iter_tag)
    eig(viennacl::matrix<double, viennacl::column_major, 1u>, viennacl::linalg::power_iter_tag)
    eig(viennacl::matrix<float, viennacl::row_major, 1u>, viennacl::linalg::power_iter_tag)
    eig(viennacl::matrix<double, viennacl::row_major, 1u>, viennacl::linalg::power_iter_tag)

I tried with power_iter_tag, too, and it works. But it's not what I need. Please advise.

Thanks for your effort! This is very promising stuff. Cheers, wolfgangp

[tsmithe]

Hi there,

thanks for the report. It looks like I had accidentally left that code path disabled during some testing earlier. I've re-enabled it now in the git master branch -- could you try that and see if it works?

I really ought to have a unit test for these bits; I'm surprised I missed it!

Cheers,

Toby

[wolfgangp]

Hi Toby, It's working fine now. wolfgangp

[tsmithe]

Great. I'll include the fix (and a unit test!) in the next release; for now, keep using the git version (it won't change much, if at all, over the next few months), and do report any more issues you encounter.

[tsmithe]

Hi wolfgangp -- just FYI: I realised that I had also failed to enable eigenvalue computations for sparse (compressed/coordinate) matrices, and I've now fixed that, too. If you need that functionality, you'll have to pull from git again!