lucfra / lds-gnn Goto Github PK

Hi!
I think the work is typically great and interesting!
But i wonder if there is a pytorch version?

Hi, thanks for this nice implementation and paper.
From my understanding, LDS-GNN updates graphs and GCN parameters at the same time for a node classification problem. So, say if in another task, the object is to find the label of the edge ( while there's no label for vertices), can LDS-GNN still fit into this task?

Many thanks.

hi, I just try to run this project, but after running python setup.py install and start, this error was appended. I am sure I have installed the GCN by pip tool.
thank you very much.

Hello! Thanks for your excellent work!
I have a question about your calculation of outer loss on validation set.
Under the standard split, the validation set for cora, citeseer and pubmed is several times larger than the training set, and it seems unfair that you compute the loss on the validation set.
Could you please let me know if I'm missing something?

Hi, i think your code is pretty good, but i wonder if there is an gpu-version?

Hi @lucfra

I tried to run LDS on the dataset Pubmed, which has 19,717 nodes and 44,338 edges. The dense adjacent matrix should occupy about 19717 * 19717 * 32 / 8 / 10e9 = 1.56 GB memory.

When we directly train a GCN model on Pubmed, the process occupies about 684 MB GPU memory (since the adjacent matrix is sparse).

On the other hand, when we train a GCN model in an LDS manner, the process occupies 30.95 GB GPU memory.

Could you provide some space complexity analysis of LDS? I did not found it in the paper. Thanks!

This is such an interesting idea and nicely done work. Thanks a lot!

Is there a GPU implementation of your algorithm? I was testing digits example and it takes a lot of time.

Hi,

Thank you for sharing the code of this awesome work!
I have one question about the kNN-GCN baseline method mentioned in the paper. Since the kNN graph is directed, I was wondering what is your strategy to apply GCNs to the directed kNN-graph. Specifically, how did you normalize the non-symmetric adjacency matrix?

Thanks!

Hi @lucfra ,

Thanks for your great work!

I ran the following code

python lds.py -m lds -e 100 -d cora

and only got about 70% test accuracy after the grid search. However, the reported performance in Figure 2 of your paper is higher than 80% with very low variance.

I did not modify any code in this repo. Can you give me some advice to reproduce the results?

Dear Author:

Great Work! I would like to run fma experiment,may I get this dataset.

Regards

Hi, i think your code is very innovation, and i try to modify it with adding lstm to the gcn model, which is in models.py, but it turns back an error "TypeError: Second-order gradient for while loops not supported.". So, could you give me some suggestions>