How can we get the difference between the node features based on the edge index in a graph classification task?

[arake001]

Suppose we have 10 nodes in a graph. Based on the edge_index, I want the difference between the node features attributes.

Something like sub_node_features = w(hpt - hjt). Here p node is the neighboring node for node j from edge index. t represents the number of node feature attributes.

[rusty1s]

Do you mean something like:

out = x[edge_index[0]] - x[edge_index[1]]