MIRALab-USTC / KGE-HAKE

The code of paper Learning Hierarchy-Aware Knowledge Graph Embeddings for Link Prediction. Zhanqiu Zhang, Jianyu Cai, Yongdong Zhang, Jie Wang. AAAI 2020.
https://arxiv.org/abs/1911.09419
MIT License
251 stars 52 forks source link

Number of parameters #5

Closed Demirrr closed 4 years ago

Demirrr commented 4 years ago

Hello,

I was wondering where one could find the number of parameters of the ModE and HAKE on WN18RR, FB15k-237 and YAGO3-10.

Cheers

zhanqiuzhang commented 4 years ago

Hi,

The number of parameters can be calculated by d_e E+d_r R, where d_e is the entity embedding dimension, E is the number of entities, d_r is the relation embedding dimension, and R is the number of relations. You can find the values of d_e, d_r, E, and R in README and our paper.

Also, you can download the codes and initialize the model with Pytorch, and then print the number of parameters (How do I check the number of parameters of a model?).

Regards

Demirrr commented 4 years ago

Life would more beautiful if everyone answers questions as precise as @zhanqiuzhang :). Appreciated!

However, Table 3 buffles me as I fail to find Hit@N and MRR performances of RotatE on YAGO3-10 in the mentioned papers. Nguyen et al. (2018), Sun et al. (2019) or Dettmers et al. (2018) do not report evaluation of RotatE in terms of MRR and H@N on YAGO3-10, do they ?

zhanqiuzhang commented 4 years ago

You can find the results on page 14 of Sun et al. (2019).

Demirrr commented 4 years ago

Oh I see, results are provided in the appendix.