run code on PROTEINS dataset error

[ColdenChan]

Hi, thanks for your code. But when I run python main.py --dataset PROTEINS, there is an error as follows:

Traceback (most recent call last):
  File "main.py", line 99, in <module>
    test_acc,test_loss = test(model,test_loader)
  File "main.py", line 65, in test
    out = model(data)
  File "/home/colden/.conda/envs/tg/lib/python3.6/site-packages/torch/nn/modules/module.py", line 493, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/colden/SAGPool/networks.py", line 45, in forward
    x, edge_index, _, batch, _ = self.pool3(x, edge_index, None, batch)
  File "/home/colden/.conda/envs/tg/lib/python3.6/site-packages/torch/nn/modules/module.py", line 493, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/colden/SAGPool/layers.py", line 20, in forward
    perm = topk(score, self.ratio, batch)
  File "/home/colden/.conda/envs/tg/lib/python3.6/site-packages/torch_geometric/nn/pool/topk_pool.py", line 18, in topk
    num_nodes = scatter_add(batch.new_ones(x.size(0)), batch, dim=0)
IndexError: dimension specified as 0 but tensor has no dimensions```

Can you help me to fix the bug? Thanks very much.

[ColdenChan]

Hi, thanks for your code. But when I run python main.py --dataset PROTEINS, there is an error as follows:
Validation loss:0.513576095168655 accuracy:0.7567567567567568 Traceback (most recent call last): File "main.py", line 99, in <module> test_acc,test_loss = test(model,test_loader) File "main.py", line 65, in test out = model(data) File "/home/colden/.conda/envs/tg/lib/python3.6/site-packages/torch/nn/modules/module.py", line 493, in __call__ result = self.forward(*input, **kwargs) File "/home/colden/SAGPool/networks.py", line 45, in forward x, edge_index, _, batch, _ = self.pool3(x, edge_index, None, batch) File "/home/colden/.conda/envs/tg/lib/python3.6/site-packages/torch/nn/modules/module.py", line 493, in __call__ result = self.forward(*input, **kwargs) File "/home/colden/SAGPool/layers.py", line 20, in forward perm = topk(score, self.ratio, batch) File "/home/colden/.conda/envs/tg/lib/python3.6/site-packages/torch_geometric/nn/pool/topk_pool.py", line 18, in topk num_nodes = scatter_add(batch.new_ones(x.size(0)), batch, dim=0) IndexError: dimension specified as 0 but tensor has no dimensions
Can you help me to fix the bug? Thanks very much.

When I set testdata batchsize=2 (test_loader = DataLoader(test_set,batch_size=2,shuffle=False)), it will be fine. However, when I run then code in dataset ENZYMES, it report the following error:

/opt/conda/conda-bld/pytorch_1565272271120/work/aten/src/ATen/native/cuda/IndexKernel.cu:60: lambda [](int)->auto::operator()(int)->auto: block: [150,0,0], thread: [58,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && "index out of bounds"` failed.
/opt/conda/conda-bld/pytorch_1565272271120/work/aten/src/ATen/native/cuda/IndexKernel.cu:60: lambda [](int)->auto::operator()(int)->auto: block: [150,0,0], thread: [59,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && "index out of bounds"` failed.
/opt/conda/conda-bld/pytorch_1565272271120/work/aten/src/ATen/native/cuda/IndexKernel.cu:60: lambda [](int)->auto::operator()(int)->auto: block: [150,0,0], thread: [60,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && "index out of bounds"` failed.
/opt/conda/conda-bld/pytorch_1565272271120/work/aten/src/ATen/native/cuda/IndexKernel.cu:60: lambda [](int)->auto::operator()(int)->auto: block: [150,0,0], thread: [61,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && "index out of bounds"` failed.
/opt/conda/conda-bld/pytorch_1565272271120/work/aten/src/ATen/native/cuda/IndexKernel.cu:60: lambda [](int)->auto::operator()(int)->auto: block: [150,0,0], thread: [62,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && "index out of bounds"` failed.
/opt/conda/conda-bld/pytorch_1565272271120/work/aten/src/ATen/native/cuda/IndexKernel.cu:60: lambda [](int)->auto::operator()(int)->auto: block: [150,0,0], thread: [63,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && "index out of bounds"` failed.
Traceback (most recent call last):
  File "main.py", line 79, in <module>
    out = model(data)
  File "/usr/local/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 547, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/colden/SAGPool/networks.py", line 45, in forward
    x, edge_index, _, batch, _ = self.pool3(x, edge_index, None, batch)
  File "/usr/local/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 547, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/colden/SAGPool/layers.py", line 24, in forward
    edge_index, edge_attr, perm, num_nodes=score.size(0))
  File "/usr/local/anaconda3/lib/python3.7/site-packages/torch_geometric/nn/pool/topk_pool.py", line 59, in filter_adj
    row, col = row[mask], col[mask]
RuntimeError: copy_if failed to synchronize: device-side assert triggered

I am not sure how to fix this bug.

[ThyrixYang]

I encountered the same error.
It seems an error related to topk in pytorch-geometric.

I opened an issue at pytorch-geometric repo here

pyg-team/pytorch_geometric#775

[rusty1s]

I identified the issue in the SAGPool layers.py code:

score = self.score_layer(x, edge_index).squeeze()

will actually squeeze the tensor to a 0-dimensional tensor if the input has shape [1, 1]. Replacing this line with

score = self.score_layer(x, edge_index).squeeze(-1)

fixes the error.

[ColdenChan]

I identified the issue in the SAGPool layers.py code:

score = self.score_layer(x, edge_index).squeeze()

will actually squeeze the tensor to a 0-dimensional tensor if the input has shape [1, 1]. Replacing this line with

score = self.score_layer(x, edge_index).squeeze(-1)

fixes the error.

Thank you very much!