Which representation is actually used

[Zhang-Each]

[Zhang-Each]

Nice work! But I have a question to know. According to the paper and code there are three representations for each graph, which one is actually used for label prediction? The causal one or the intervention representation?

[Zhang-Each]

If the intervention representation is used, how to make the prediction stable due to the random addition?

[Namkyeong]

I'm also wonder about this. Which prediction is used for performance?

[czstudio]

After reviewing the paper and code, it appears that the intervention representation is used for label prediction. The paper specifically states that the intervention representation is used for both training and evaluation of the model (Section 3.1 and Section 3.2). Additionally, the function generate_intervention_data in the code is used to create the intervention representation of the graph, which is then fed into the model for label prediction (Section 4.1 and Section 4.2).