I want to use GIN implementation of DGL for graph regression. I know I should change loss function to MSE loss and output_dim ==1.
my questions are
- should I add one more layer at the end of the GIN structure with output dimension =1?
- how should I change the gindt.py code specifically for this line self.gclasses = len(self.glabel_dict)
I’m not sure how to use a regression target for every graph and glabel_dict at the same time.