mikekatz04 / lisaanalysistools Goto Github PK

Some utilities in lisatools/diagnostic.py support GPU usage (i.e. use_gpu=True) but not consistently, leading to type errors where numpy is used instead of cupy.

One example of this is when computing an overlap (which involves normalising the inner product), inner_product() is recalled within the first inner_product() call but GPU preference is not specified, leading to a type error.

Fisher matrix generation also needs to know about whether GPUs are being used or not, as the waveforms used to compute the FIM may be generated with a GPU which is currently fed into numpy code resulting in type errors.

I will submit a pull request soon for these changes.

The LSA/FIM validity check from Vallisneri (2008), implemented in diagnostic.py as mismatch_criterion(...) is not correct.

Line 431, vec_delta = x[l] * v[:, l] / np.sqrt(w[l]), does not take a linear combination of the eigenvectors weighted by eigenvalues, which would be something like vec_delta[i] = np.sum(x * v[i,:] / np.sqrt(w)). If one were also to write vec_delta += .... that would fix it too.

I will set up a pull request.