Suchun-sv
commented
11 months ago Thanks for your interests of our work, I repeat the code and got non-nan results, could you please replay your code with simple adjustments (e.g., changing your random seed). Feeling free for providing your env by running "pip freeze" for us debugging?
Based on the given codes, datasets and parameter settings. I ran 'python train_SGDD.py --dataset cora --nlayers=2 --beta 0.1 --r=0.5 --gpu_id=0' but the loss_avg and the loss of training gcn model became 'nan' after training the synthetic data for 450 epochs. Do I miss some important Steps? The following is my training log. cora.txt