framework to apply convolutional neural networks to neuroimaging surfaces
#Dependencies:
standard python packages: pip install numpy argparse nibabel tensorflow networkx
python package for gcn implementation:
git clone https://github.com/tkipf/gcn
Make sure you can run:
python train.py
If you have trouble, open up utils.py
For us, sp.diags(r_inv) needed to be changed to sp.diags(r_inv, offsets=0) (2 instances of sp.diags to change in total)
For reading in neuroimaging data we're going to be using io_mesh.py
It comes from laminar_python, which has some awesome functions so please do download the whole package (package: https://github.com/juhuntenburg/laminar_python paper: https://doi.org/10.3897/rio.3.e12346). but we're currently only using io_mesh.py so I've copied it to this package.
git clone https://github.com/kwagstyl/convolutional_meshes
add paths:
export PYTHONPATH=$PYTHONPATH:/path/to/convolutional_meshes