One qusetion about learned_graph.

[ShiinoMashiro]

Thanks for your excellent work,
According to your code , I want to know if the X_data in the same batch share the same learned graph ?

[d-ailin]

Thanks for your interest.
If you indicate the adjacency matrix A in Eq(3), then it is yes. The graph is the same for all input data. But when it comes to the attention part in Eq(6)-(8), each sample can get different attention weights as the attentions are computed based on the each sample and the global sensor embeddings.

[ShiinoMashiro]

Thank you , after changing the outlayer and dataset , can the model with graph_layer also be used for graph classification tasks ?

[d-ailin]

Potentially yes. I think it depends on your task. You could use the embedding before the outlayer potentially for any downstream task when the learned representations have some good properties, such as clustering.