Related to #304. It looks like recent versions of nilearn now raise errors on transformed data that isn't nsamples x nvoxels. So whenever we run brain.ttest(threshold_dict=...) plotting and saving (to nifti) the thresholded t map produces TypeErrors, like
~/anaconda3/lib/python3.8/site-packages/nilearn/masking.py in _unmask_3d(X, mask, order)
831 n_features = mask.sum()
832 if X.shape[0] != n_features:
--> 833 raise TypeError('X must be of shape (samples, %d).' % n_features)
834
835 data = np.zeros(
TypeError: X must be of shape (samples, 51029).
@ljchang This is another example where we should prioritize revisiting #304 I'm not seeing an advantage of changing the shape of a brain data object ever, as oppose to just setting voxels to 0 whenever we threshold. And this new error supports the opting for the latter approach across our toolbox.
Related to #304. It looks like recent versions of nilearn now raise errors on transformed data that isn't
nsamples x nvoxels
. So whenever we runbrain.ttest(threshold_dict=...)
plotting and saving (to nifti) the thresholded t map produces TypeErrors, like@ljchang This is another example where we should prioritize revisiting #304 I'm not seeing an advantage of changing the shape of a brain data object ever, as oppose to just setting voxels to 0 whenever we threshold. And this new error supports the opting for the latter approach across our toolbox.