Runlin Lei

Hi! I've encountered the same problem. Metattack works fine when using the following environment on ubuntu 16.04.12 ``` numpy==1.18.1 scipy==1.6.2 torch==1.8.1 torch_geometric==1.6.3 torch_scatter==2.0.9 torch_sparse==0.6.12 ``` But it shows that CUDA...

Hi! I write an issue in #128, which seems to be helpful in avoiding OOM in MetaAttack. I didn't check the solution thoroughly on different devices. Please let me know...

Hi! I'm sorry that I don't know why the changes don't work in your environment. Here, I will provide my environment and detailed steps that work for me: ``` numpy==1.23.4...

Hi! The phenomenon is quite similar to me when no change is made. The OOM occurs in the middle of training. For you to debug, I suggest that you first...

Hi! You can try MetaApprox from [mettack.py](https://github.com/DSE-MSU/DeepRobust/blob/master/deeprobust/graph/global_attack/mettack.py). It is an approximated version of Metattack. In [ProGNN](https://arxiv.org/abs/2005.10203), PubMed is attacked using MetaApprox. If it still does not work for you, maybe...

Thanks! I was confused before by the reply in issue[ #12 ](https://github.com/ChandlerBang/Pro-GNN/issues/12). You only mentioned changes in the sequential attack in [test.py](https://github.com/DSE-MSU/DeepRobust/blob/master/examples/graph/test_nettack.py), but not the changes in node selection. It...

Hi. Thanks for your detailed answer. In my experiments, multiplying `lr` with `num_budgets` also improves the performance of CE loss, especially in the poison setting, but is relatively less useful...