Regarding the issue of the testing model about samgc HOT 3 CLOSED

Could you please provide more details about your reproduction process? In general, the results might be vary slightly when training on different machines with different environments. However, testing results with *.pkl should be consistent with the reported ones.

from samgc.

The entire code, I only modified the line model = MultiGraphAutoEncoder(feat_dim, hidden_dim, latent_dim, class_num, lam_emd=lam_emd, order=order, view_num=graph_num), removing the lam_emd parameter because keeping it would result in a TypeError: init() got an unexpected keyword argument 'lam_emd'. Then, I proceeded to evaluate using samgc_acm_for_test.pkl, and the obtained results are as follows:

from samgc.

Hi,

Thanks for your reply and detailed information.

The issue is arisen due to the version difference. I uploaded the latest version, and hope the new version can address your concern. Please note that use the hyperparameters shown in test.sh for testing. You can test SAMGC via sh test.sh or python samgc.py --dataset 'acm' --model_name 'samgc_acm_for_test.pkl' --order 16 --hidden_dim 256 --latent_dim 64 --use_cuda True --cuda_device 0 to run the model.

Again, we appreciate your issues, and feel free to contact us if you have any further questions in reproducing SAMGC.

Best regards.

from samgc.