graph-com / ed-hnn Goto Github PK

I just follow the readme to perform one experiment on cpu:
python train.py --method EDGNN --dname senate-committees-100 --All_num_layers 8 --MLP_num_layers 2 --MLP2_num_layers 2
--MLP3_num_layers 2 --Classifier_num_layers 2 --MLP_hidden 512 --Classifier_hidden 256 --aggregate mean
--restart_alpha 0.5 --lr 0.001 --wd 0 --epochs 500 --runs 10 --feature_noise 1.0
--data_dir <data_path> --raw_data_dir <raw_data_path>

It returns the results:
Highest Train: 100.00 ± 0.00
Highest Valid: 77.29 ± 4.18
Final Train: 97.16 ± 2.36
Final Test: 68.73 ± 3.56

my question is why it is much higher than the value 64.79 ± 5.14 from paper?
though I did do it on GPU but it differ a lot, anything I miss?

Hi,

Thanks for sharing the codes. This paper is very interesting.

When I try to reproduce the results and run 'python train.py --method HCHA --dname senate-committees-100 --All_num_layers 8 --MLP_num_layers 2 --MLP2_num_layers 2 --MLP3_num_layers 2 --Classifier_num_layers 2 --MLP_hidden 512 --Classifier_hidden 256 --aggregate mean --restart_alpha 0.5 --lr 0.001 --wd 0 --epochs 500 --runs 1 --feature_noise 1.0 --cuda 0 --data_dir data/ --raw_data_dir raw_data/senate-committees', I get an error.

Error: 'ValueError: Encountered tensor with size 597 in dimension 0, but expected size 282.'. in line 161, models/hcha.py. Could you give me some suggestions on how to solve this problem if you are available? Thanks in advance.

Cheers