ENH: add average error to ErrorMap #1039
Conversation
|
@satra Thanks for the comments. I will update the changes file and make some according tests. |
|
@oesteban - in the following chunk of code in ErrorMap, were you just trying to calculate the euclidean distance voxel-wisely ( double checking this because I am making some tests for it -- Thanks in advance |
|
@lsqshr you are right, but if I recall it well, I use The |
|
perhaps we can also add the frobenius norm here and keep expanding the list of metrics. |
…euclidean average `errvectorexp = np.zeros_like(mskvector)` used to generate a uint8 vector by default which loses the precision when calculating the special volumes with float voxels of small values like the jacobian determinant volume
lsqshr
changed the title
|
Just made the update and some tests with dummy |
| mskvector = msk.reshape(-1) | ||
| msk_idxs = np.where(mskvector==1) | ||
| refvector = ref_data.reshape(-1,comps)[msk_idxs].astype(np.float32) | ||
| tstvector = tst_data.reshape(-1,comps)[msk_idxs].astype(np.float32) | ||
| diffvector = (refvector-tstvector) | ||
|
|
||
| # Scale the diffrernce |
There was a problem hiding this comment.
Minimal typo (should read difference).
…_2_errormap Conflicts: CHANGES
|
@oesteban, thanks for reviewing the code and the comments. I have just made the according fixes and synced this branch with the upstream. |
ENH: add average error to ErrorMap
Add an average distance output to the
nipype.algorithms.metrics.ErrorMap, which simpy get the square root average/the sum average of the errormap vector.