I noticed that LIRA's C code would occasionally sample NaN values, which then affects the computation of the posterior mean. However recomputing the posterior mean from the image trace using np.nanmean just works fine. ideally we can fix the cause of NaN values being generated. If it is expected, one should adapt at least the posterior mean computation.
I noticed that LIRA's C code would occasionally sample NaN values, which then affects the computation of the posterior mean. However recomputing the posterior mean from the image trace using
np.nanmeanjust works fine. ideally we can fix the cause of NaN values being generated. If it is expected, one should adapt at least the posterior mean computation.