Hi,
I am in the process of looking at the GraphWaveNet model and its output information. I am trying to use Captum's integrated gradient analysis to understand each feature's attribute weight on nodes.
Has anyone done this with Graph WaveNet and if so Please let me know. I am struggling becuase the model.py changes the dimensions from 3D to 4D at several locations and IG does not like this at all. I've tried to use a wrapper to get around but no luck.
If you happen to have figured this out please let me know what you did to the repo code. or in you explaining IG code.
Thank you!
Best,
Gabe