[ ] Pass the timesteps to the adjacency learner when you pass them to the classifier.
i.e.
I retrieve batch (x, y) from the dataloader.
I pass (x) to the adjacency learner, and retrieve an adjacency matrix (A)
I then use (x, A, y) in my classifier to predict VCI.
The intuition is that A can depend on the timestep. So you a model which outputs an adjacency matrix per training-instance.
Passing all the timesteps seems nuts.
i.e. I retrieve batch (x, y) from the dataloader. I pass (x) to the adjacency learner, and retrieve an adjacency matrix (A) I then use (x, A, y) in my classifier to predict VCI.
The intuition is that A can depend on the timestep. So you a model which outputs an adjacency matrix per training-instance.