split method of train.txt and facts.txt

Question

Closed this issue 2 years ago · 1 comments

Hi,
Your work is great. I'm reproducing your paper recently, but i haven't found how you divide train.txt and facts.txt in your paper. Is it Random?

Answer 1 · 2022-05-24T03:51:56.000Z

Hi,

The splits in this repo follow Neural-LP: https://github.com/fanyangxyz/Neural-LP/tree/master/datasets. I think this is a random split.

In our code https://github.com/AutoML-Research/RED-GNN/blob/main/transductive/load_data.py#L152, we provide a shuffle function since we find that randomly shuffling the set of facts and trains can slightly improve the performance.