It appears that the nnls() method is reimplemented in SciPy 1.12, and due to some changes in the termination criteria, inversion_with_raytransfer.py, inversion_with_voxels.py bolometer demos and admit_tomographic_inversion.py from #427 return zero emissivity profiles.
Normalising the c_matrix and d_vector by d_vector.max() before passing them to nnls(), solves the problem.
Also, the new nnls implementation is much slower if atol is not set. I suggest adding **kwargs to invert_regularised_nnls() so that the user can control maxiter and atol parameters.
vsnever
commented
2 months ago
It appears that the
nnls()method is reimplemented in SciPy 1.12, and due to some changes in the termination criteria,inversion_with_raytransfer.py,inversion_with_voxels.pybolometer demos andadmit_tomographic_inversion.pyfrom #427 return zero emissivity profiles.Normalising the
c_matrixandd_vectorbyd_vector.max()before passing them tonnls(), solves the problem.Also, the new
nnlsimplementation is much slower ifatolis not set. I suggest adding**kwargstoinvert_regularised_nnls()so that the user can controlmaxiterandatolparameters.