Static graph for whole dataset

[ralfriegel5]

I want to solve a regression task with n samples. The inputs are graph-structured. The graph does not change over the samples, i.e. it is a static graph for the whole dataset. In other words, only the node features change over the samples.

As each sample is supposed to be one Data instance, putting the graph into each Data sample seems to be inefficient and redundant. Is there a more efficient way to do this?

Many thanks!

[rusty1s]

You can also define your graph outside of data objects, e.g.

class Dataset(...):
     def __init__(self, ...):
          self.edge_index = ...
          self.data, self.slices = torch.load(self.processed_paths[0])
      def __getitem__(self, idx):
          data = super().__getitem__(self, idx)
          data.edge_index = self.edge_index
          return data

[ralfriegel5]

very simple. great, thanks