search

tadpole / DHNE

the Implementation of "Structural Deep Embedding for Hyper-Networks"（AAA20I8）
http://nrl.thumedialab.com/Structural-Deep-Embedding-for-Hyper-Networks
44 stars 14 forks source link

关于输入数据 #5

Closed Theheavens closed 4 years ago

Theheavens commented 4 years ago

您好，输入的三元组是怎么抽的，是把网络中所有的三元组都抽出来，还是像random_walk一样抽样，可以重复，且数量由超参决定

tadpole commented 4 years ago

一个epoch是所有的三元组，然后随机采样同等数量的负边，具体实现参见dataset.py中的next_batch函数。

https://github.com/tadpole/DHNE/blob/2057b24726468794aa2da01df08bf757dbbb6614/src/dataset.py#L27