inconsistency in model.py predict() with timesteps for subject vs object

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

inconsistency in model.py predict() with timesteps for subject vs object #64

Open JuliaGast opened 2 years ago

JuliaGast commented 2 years ago

Hello,

I found an inconsistency: predict(), l. 223: _, sub, prob_sub = global_model.predict(self.latest_time, self.graph_dict, subject=True)

predict(), l. 261: _, ob, prob_ob = global_model.predict(t, self.graph_dict, subject=False)

why do you pass the self.latest_time for the subjects, and t for the objects, whereas clearly they are not the same (see line 222)?

woojeongjin commented 2 years ago

Each subject and object have different latest times since each subject's last interaction (with possibly another object) and each object's last interaction (with possibly another subject) are different.