I initially posted on neurostars about this: https://neurostars.org/t/nipype-merging-outputs-from-bedpostx-parallel-fails/31733/1. But I think I have a quick fix, so reporting it here. Will also do PR.

As reported in the neurostars post, I am trying to run bedpostx_parallel workflow in old (?) nipype. It works fine during all of my test runs on downsampled data (and even some previous runs on full res data).

However, when I recently tried to run it on the full res data, it failed at the step where it tried to merge all the outputs from parallel nodes into single files.

I looked into the issue, and it seems that on this line:

the output of np.squeeze is a 0D array, and getting the length of 0D array in a subsequent line isn’t possible.

I could just be an anomaly due to the heuristic nature of the algorithm, and it would probably not be an issue on a subsequent run. But I still think that it could be handled within nipype.

Here's the error traceback: