search

Yanqi-Chen / GWNN

Course project. A implementation of Graph Wavelet Neural Network (ICLR 2019)
GNU General Public License v3.0
10 stars 3 forks source link

Error on Pubmed #2

Open abhigoku10 opened 3 years ago

abhigoku10 commented 3 years ago

@Yanqi-Chen thanskfor sharing the code base when i select pubmed i get the following error "output= torch.mm(torch.mm(self.wavelets,torch.diag(self.filter)),torch.mm(self.wavelets_inv,transformed_features)) RuntimeError: CUDA out of memory. Tried to allocate 1.45 GiB (GPU 0; 6.00 GiB total capacity; 4.39 GiB already allocated; 259.94 MiB free; 4.41 GiB reserved in total by PyTorch)" Thanks in advance

Yanqi-Chen commented 3 years ago

It seems that your GPU memory size (6GB) is insufficient for this code. You may need a GPU with a larger memory size.

abhigoku10 commented 3 years ago

@Yanqi-Chen but other datasets r running successfully with large batch size also

Yanqi-Chen commented 3 years ago

@Yanqi-Chen but other datasets r running successfully with large batch size also

That's right, but Pubmed is much larger than the others.

Items Cora Citeseer PubMed
#Nodes 2708 3327 19717
#Edges 5429 4732 44338
#Features 1433 3703 500
#Classes 7 6 3
abhigoku10 commented 3 years ago

what is the accuracy ur gettting for pubmed data

Yanqi-Chen commented 3 years ago

what is the accuracy ur gettting for pubmed data

76.8±1.1%

abhigoku10 commented 3 years ago

@Yanqi-Chen thanks for cora and citseer i am getting an accuracy of 76.90 and 49.20 is it right ??

Yanqi-Chen commented 3 years ago

To get the result close to the claim of the ICLR paper, omit --fast. The acc. on test set should be like: image Orange is Cora: 81.3±2.1% Blue is CiteSeer: 71.1±1.5%