wangxuhongcn / ocgnn Goto Github PK

Hi, I am wondering why you set model.train() outside of the training loop? And I see that during the training loop, you seem to use the function fixed_graph_evaluate(), which sets model.eval(). Can I take that to mean the GNN model is at eval status during that time? Thanks.

Hi, I am curious about how to use the OCGNN to do anomaly detection on my own datasets. Is there any tips which I should know?

Hi Xuhong, I’m a junior from JLU trying to run your code.
The problem is, the code is not working on my PC having error showing that the tensors seems are arranged on different device. I’ve changed the computer but the problem is the same. I was running the main.py by typing “ python main.py —dataset Cora —module GCN —nu 0.1 —lr 0.001 —h-hidden 32”
All the packages required are downloaded, and the data is loaded and preprocessed. The error content are demonstrated in the title. Even if I change the computer, the issue still exist. BTW, the gpu is Titan RTX. Could you pls take a minute and check this issue?
Looking forward to your reply.

Hi Xuhong,

I hope you're doing well. I wanted to seek help with the below error which I'm facing while running main.py file:

Error: "FileNotFoundError: [Errno 2] No such file or directory: '/floyd/home/log/cora+OC-GAE.log'"

This is what I'm trying to run: python main.py --dataset cora --module GAE --gpu 0 --n-epochs 5000 --seed 46

Looking like the path given to read the log file isn't correct because of which it shows no such file or directory. Can you check if you are also able to replicate this bug?

Best Regards,
Vatsal Patel

I'm facing another error that I wanted to bring to your attention. I'm working on this error too but wanted to drop a message to see if you know what's happening here.

ERROR: cannot find key: --dataset

what's the remove_edges_from function and add_edges_from function in datasets/dataloader.py? I didn't find it in DGL, and it seems not a property of sub-class

Hi Xuhong,

Thanks for the great contribution.
I have a question about the GNN layers that can be used in the OCGNN framework. I just realized the GNN layers proposed in your paper/code are semi-supervised/unsupervised. Is it possible to replace the GNN layer with a supervised approach? In other words, is the semi-supervised/unsupervised structure required in OCGNN?
Look forward to your replying.

Trying to use OCGNN on custom dataset containing node features and adjacency matrices. Any helper functions that support OCGNN for custom dataset? Thanks!