(The expected output of np.sum(np.isnan(scipy.linalg.svd(matrix, full_matrices=False)[0])) is 0; the bug is that the first call to scipy.linalg.svd returns a U matrix with a column of nans (all in the ...