having problems when running the demo - about 'torch-sparse'

[CarbonBlaack]

• Try my best to search on the internet.
• Browse all the issues in this project

I try to run the demo and follow the instruction here(based on Installation 1-4 steps).

conda create -n stellar python=3.8
source activate stellar
conda install pytorch cudatoolkit=11.3 -c pytorch
conda install pyg -c pyg
pip install -r requirements.txt

I also add matplotlib==3.6 to the requirements.txt according to another issue. in this project.

Setup specs:(Ubuntu 20.04)

python                    3.8.17               h955ad1f_0
pytorch                   1.12.1          cpu_py38h9dbd814_1
pyg                       2.3.1           py38_torch_1.12.0_cu113    pyg

But when I run the demo, there is an error: "ImportError: 'SparseTensor' requires 'torch-sparse'"
The specific messages are here:

ImportError                               Traceback (most recent call last)
Cell In[7], line 2
      1 stellar = STELLAR(args, dataset)
----> 2 stellar.train()
      3 _, results = stellar.pred()

File [~/Projects/2023_08/stellar/STELLAR.py:176](https://vscode-remote+ssh-002dremote-002b166-002e111-002e130-002e162.vscode-resource.vscode-cdn.net/home/wangzian/Projects/2023_08/stellar/~/Projects/2023_08/stellar/STELLAR.py:176), in STELLAR.train(self)
    174 seed_optimizer = optim.Adam(seed_model.parameters(), lr=1e-3, weight_decay=5e-2)
    175 for epoch in range(20):
--> 176     self.train_supervised(self.args, seed_model, self.args.device, self.dataset, seed_optimizer, epoch)
    177 novel_label_seeds = self.est_seeds(self.args, seed_model, self.args.device, self.dataset, clusters, self.args.num_seed_class)
    178 self.dataset.unlabeled_data.novel_label_seeds = torch.tensor(novel_label_seeds)

File [~/Projects/2023_08/stellar/STELLAR.py:29](https://vscode-remote+ssh-002dremote-002b166-002e111-002e130-002e162.vscode-resource.vscode-cdn.net/home/wangzian/Projects/2023_08/stellar/~/Projects/2023_08/stellar/STELLAR.py:29), in STELLAR.train_supervised(self, args, model, device, dataset, optimizer, epoch)
     26 sum_loss = 0
     28 labeled_graph = dataset.labeled_data
---> 29 labeled_data = ClusterData(labeled_graph, num_parts=100, recursive=False)
     30 labeled_loader = ClusterLoader(labeled_data, batch_size=1, shuffle=True,
     31                             num_workers=1)
     33 for batch_idx, labeled_x in enumerate(labeled_loader):

File [~/anaconda3/envs/stellar/lib/python3.8/site-packages/torch_geometric/deprecation.py:23](https://vscode-remote+ssh-002dremote-002b166-002e111-002e130-002e162.vscode-resource.vscode-cdn.net/home/wangzian/Projects/2023_08/stellar/~/anaconda3/envs/stellar/lib/python3.8/site-packages/torch_geometric/deprecation.py:23), in deprecated..decorator..wrapper(*args, **kwargs)
     21     out += f", {details}"
     22 warnings.warn(out)
...
File [~/anaconda3/envs/stellar/lib/python3.8/site-packages/torch_geometric/typing.py:49](https://vscode-remote+ssh-002dremote-002b166-002e111-002e130-002e162.vscode-resource.vscode-cdn.net/home/wangzian/Projects/2023_08/stellar/~/anaconda3/envs/stellar/lib/python3.8/site-packages/torch_geometric/typing.py:49), in SparseTensor.__init__(self, *args, **kwargs)
     48 def __init__(self, *args, **kwargs):
---> 49     raise ImportError("'SparseTensor' requires 'torch-sparse'")

ImportError: 'SparseTensor' requires 'torch-sparse'

After that, I have already tried the following:

pip install torch_geometric

and

pip install torch_sparse

but it still doesn't work.

torch-sparse              0.6.17                   pypi_0    pypi

Would anyone be able to help me?

[JLrumberger]

I had the problem before, then re-installed with with the following versions on linux and it works now:

conda create -n stellar python=3.8
conda activate stellar
conda install pytorch==2.0.0 pytorch-cuda=11.8 -c pytorch -c nvidia
conda install pytorch-scatter pytorch-sparse -c pyg
python -m pip install 'matplotlib<3.7'

[LukasHats]

@JLrumberger Just curious, did you get STELLAR to run? How long did it take? Hinting towards here

[JLrumberger]

Yes, I got it to run and it was pretty fast as far as I remember. So I got results within minutes on my datasets.