I’m out of my depth here and could use some guidance. I have a dataset consisting of multiple files where the data nicely fits into a graph structure. The data can also be segmented in a time-wise manner.
I was considering use graph convolutional layers combined with an RNN variant in PyTorch and PyTorch Geometric to hopefully capture spatial and temporal features. The ultimate goal is graph classification.
I’ve already created a homogeneous multigraph in a geometric data object per file. In this graph, only edges with a specific feature value are directional, which introduces a sense of sequence. All other edges are bidirectional, representing relationships where the direction doesn’t matter. I also noted down the timestamps of the nodes in the time attribute.
Is my current data modeling approach sufficient? What additional considerations should I be aware of? I’m aware of geometric_temporal, but that seem more focused on analyzing a single time series, while my case involves dealing with one temporal signal per file.
Any insights are welcome.