ZHANG YONGQI

Hi, thanks for your interest. The implementation of triplet classification task is quite easy. You can refer to the code in OpenKE https://github.com/thunlp/OpenKE for testing. The training process is same...

你好，cuda应该是不支持集显的。如果需要运行，可以将代码中的.cuda()都删掉，这样可以在cpu上运行。速度可能比较慢，最好是找一张nvidia 2080ti以上的版本试运行

Hi, we did not test it on FB15K, but it can be easily done. For the facts.txt file, you can randomly pick up some samples (e.g. 1/3 or 1/4) from...

你好， Repeat Ratio统计20个epoch内，重复的负样本的比例，可以通过python dict()来记录实现。 Non Zero Loss Ratio统计20个epoch内loss 绝对值大于一个threshold（如1e-6）的比例，可通过计数实现。

Yes, it generally takes several hours. FB15k-237 is denser than the other datasets, resulting in a much larger subgraph size. Considering that 2080Ti is a bit old, 5 hours per...

时间太久，这些细节记不清了，代码主要是用来运行整个算法的。大致是推导了一个关于embedding的表达式，感兴趣的话可以自己定义一些标准来展示。

Thank you for your attention. - For the code, please refer to https://github.com/AutoML-Research/AutoSF/tree/AutoSF-OGB. - For the model details, please refer to https://arxiv.org/pdf/2107.00184.pdf.

Hello, thanks for your attention. Our inductive data follows the paper "Inductive Relation Prediction by Subgraph Reasoning" in ICML 2020. You can refer to the contents and the source code...

原理上大概是这样的，可以print出来检查下细节问题

有的，参考这里 https://github.com/yzhangee/NSCaching/blob/master/base_model.py#L269