I am trying to construct a GCN for image classification where each pixel is a single node in the graph. However I want to train and test the model within the same graph, so I constructed a single adjacency matrix for the entire dataset but now I am confused how the splitting should be done.
- Do I sample random nodes from the initial dataset and put it in test set?
- If so should I construct a different adjacency matrix for these nodes or let the previous adjacency matrix be used? How do I even do this?
- Can I use a single adjacency matrix for training and testing?
I am a beginner so please help me understand and correct my statements if I am wrong anywhere. My task is to train and test the model in a single graph
Are there any methods out there for training and testing the graph network in a better way