Open wxyxy33 opened 4 months ago
Hi, thanks for your questions. 'Wandb' is a platform to help track, visualize, and analyze experiments. We just neglected to remove a portion of the 'wandb' item, which has caused the problem. We have deleted all the item about 'wandb' in main.py.
Thank you so much for your help! I download the new codes, but a new problem occurred. In the "QueryDataset.py"file, something wrong (AttributeError: 'NoneType' object has no attribute 'shape') when running "self.mask = torch.zeros(self.y.shape)", I tried to solve this by print self.y.shape, in the terminal, we can see some outputs. But at last, it is said "Warning: self.y is None" as you can see in this image. What should I do to solve this error?
I have met the problem before. It may be due to a compatibility mismatch between the version of PyTorch or PyTorch Geometric and the CUDA version. You need to install the PyTorch and PyTorch Geometric that are compatible with your CUDA version.
Thank you for your response! I am just going to try create a new environment. Thank you again! By the way, after this model replication, I want to try this model on new datasets. I notice you mentioned in this project"And to preprocess, divide the graph into two parts, train.csv and test.csv, you can refer to citation_preprocess.py." I didn't find the citation_preprocess.py, could you give me some help about this? Again, thank you sooooooo much!
Hi, I have added the citation_preprocess.py.
Hi, thank you for your file. I already run the whole codes, but I want to ask about while there is no ego nodes, or there is no partition method in the original dataset, what method should be down? For example, I use second-order neighbors of a node and generate the subgraph for training, but the F1 score results seems not so good. I noticed in your paper, the citeseer dataset is also used. I'd appreciate it if you could let me know what you do with the citeseer dataset.
I tried to generate bigger subgraphs for training, the F1-score seems better for citeseer dataset, but still far way for your result which is around 0.68. So I still want to ask the method for citeseer dataset, and what method would you recommend for those dataset without special structure? Thank you so much for your help!!!
I have updated the util.py file by adding a 'load citation' process, which aligns with the content described in the paper. There are no additional special processes involved.
When I run the script as "readme.md", it occurs when just start training. In the main.py file, line 19 didn't give the argument wandb_run. What should I do for fixing this problem?