Closed josephdviviano closed 7 years ago
generating a test epitome subject that fails (DTI_CMH_S036_01
) so I can debug.
Looks like forcing randomized PCA on these subjects fixes the issue. Not ideal, but for our use case the difference in the algorithms seems non-significant. See b7c02442a356993e52da38b4df00893a0f18b9eb
I'm going to continue monitoring for problems, but consider this closed for now.
I found this bug that suggests my normal PCA failing (was specifically 'SVD did not converge') was due to the particular version of scipy I'm using see this answer. Confirmed my data had non nans
or infs
.
ERROR: failed to find the top principal components of input data: