search

INK-USC / RE-Net

Recurrent Event Network: Autoregressive Structure Inference over Temporal Knowledge Graphs (EMNLP 2020)
http://inklab.usc.edu/renet/
435 stars 95 forks source link

How can I set RE-NET w. GT to reproduce experiment #62

Open JuliaGast opened 2 years ago

JuliaGast commented 2 years ago

Hello, In your paper, table 2 you are showing an interesting ablation study, where you show also RE-NET w. GT. I would like to understand how, in the code, I can set this feeding of the GT.

As far as I understand, I would have to feed after each predicted timestep, the ground truth graph, instead of the predicted graph, before going to the next timestep, right?

Is there a configuration parameter to set? If yes: which one?

Or do I directly need to modify the code? If yes, I assume somewhere in model.py predict() I would have to feed the gt_graph instead of the predicted graph, is this correct? and what exactly would I have to modify?

Looking forward to your reply Kind Regards Julia

woojeongjin commented 2 years ago

Yes you should feed the ground-truth graphs instead of predicted graphs.

JuliaGast commented 2 years ago

Thank you for your reply.

I assume I would have to feed the ground truth in model.py predict() at self.graph_dict[self.latest_time.item()] instead of:

image

Is this correct? Are there any other modifications to be done?

binchen4110 commented 1 year ago

Thank you for your reply.

I assume I would have to feed the ground truth in model.py predict() at self.graph_dict[self.latest_time.item()] instead of:

image

Is this correct? Are there any other modifications to be done? The same question. Do you succeed?