search

ME-ICA / aroma

ICA-AROMA, as a Python package. A work in progress.
Apache License 2.0
7 stars 11 forks source link

Generate a benchmark dataset for integration tests #29

Closed oesteban closed 3 years ago

oesteban commented 3 years ago

I'd suggest pulling out the MELODIC node of fMRIPrep as run by CircleCI on ds005 (because this dataset is heavily downsampled, so the computational burden is not terrible).

fMRIPrep should also generate the necessary TPMs in native space.

vinferrer commented 3 years ago

Hi oscar, could you explain a little bit more? I understand that ds005 is lighter than the nilearn data we are using. Where can we find ds005?

vinferrer commented 3 years ago

This one:

wget --retry-connrefused --waitretry=5 --read-timeout=20 --timeout=15 -t 0 -q \
   -O ds005_downsampled.tar.gz "https://files.osf.io/v1/resources/fvuh8/providers/osfstorage/57f32a429ad5a101f977eb75"
vinferrer commented 3 years ago

I don't see any motion parameters, do we have to calculate them apart? Also i see no reference image, so do you use the mean of the image or something similar to get the motion parameters? Probably it would be better to run them once and have a separate osf tar file with the motion parameters already there.

oesteban commented 3 years ago

That's the "BIDS-raw" version of the dataset. We don't store any preprocessed copy.

But now that I think about it, using a downsampled dataset might not help us make the CI load lighter by a lot - but will definitely make it difficult to check the results when we move from standard space to native.

Sorry for the initial suggestions of ds005, but the nilearn data you guys were using could be just fine!

vinferrer commented 3 years ago

Should we close this issue then?

oesteban commented 3 years ago

No, but prepare the dataset based on the nilearn data you were already using.

I realize you're right. Sorry I'm really slow today.