search

SongtaoLiu0823 / LAGNN

[ICML 2022] Local Augmentation for Graph Neural Networks
66 stars 14 forks source link

When is the cvae_model saved in Citation? #1

Closed Xiuchen519 closed 2 years ago

Xiuchen519 commented 2 years ago

Hi, author! Thank you for opening the source code! According to the description of ’Active Learning Trick‘ in the appendix, for Citation dataset, I think cvae_model should be saved after calling generated_generator in cvae_pretrain.py. But I can't find the code to save it. And I find another file cvae_tune.py, it doesn't seem to work, since there is no generated_features function in the cvae_pretrain.py. I just feel uncertain about that when cvae_model should be saved for Citation dataset. Hope to get your reply! Thank you!

SongtaoLiu0823 commented 2 years ago

Hi, author! Thank you for opening the source code! According to the description of ’Active Learning Trick‘ in the appendix, for Citation dataset, I think cvae_model should be saved after calling generated_generator in cvae_pretrain.py. But I can't find the code to save it. And I find another file cvae_tune.py, it doesn't seem to work, since there is no generated_features function in the cvae_pretrain.py. I just feel uncertain about that when cvae_model should be saved for Citation dataset. Hope to get your reply! Thank you!

Hi,

We use the cvae_tune.py to tune the hyper-parameters of CVAE for citation dataset. After tuning, you can use cvae_pretrain.py to get the generator and save it.

SongtaoLiu0823 commented 1 year ago

Hi,

Apologies for the previous misunderstanding. I realized that I had not updated the function name when submitting my code to the GitHub repository. The correct function name should be "generated_generator" instead of "generated_features." I have now fixed this issue.

For each hyperparameter set, I use cvae_pretrain.py to generate features, and then use cvae_tune.py to evaluate the CVAE model using these generated features. I select the hyperparameter set with the highest accuracy on the validation set.

Indeed, in comparison to the OGB dataset, I had not previously written cvae_generate_citation.py for citation datasets. However, I have now created this script, which can be found here: cvae_generate_citation.py.

Thank you!