search

lrjconan / GRAN

Efficient Graph Generation with Graph Recurrent Attention Networks, Deep Generative Model of Graphs, Graph Neural Networks, NeurIPS 2019
MIT License
462 stars 97 forks source link

why GRAN model can be trained so fast while tested really slowly? #15

Closed xiangsheng1325 closed 3 years ago

xiangsheng1325 commented 3 years ago

Hi, I really appreciate your excellent work. Actually, I am using your default hyperparameters and I am trying to train the GRAN model on Cora citation network datasets. But I have found a weird thing that the model is trained by 5000 epochs only using 2 seconds. The more weird thing is that when I test the model, it will consume more than half an hour. The test results are not good yet. I guess that is because the dataset contains only one graph with more than 3000 nodes. So can you explain why the weird speed occurs and how can I accelerate the test process and improve the quality of generated graphs? thanks

xiangsheng1325 commented 3 years ago

I have figured out the reasons. It is because the hyper-parameter 'num_subgraph_batch' is not set to 1.