Dataloader for Bipartite Data

[stefansei]

Hi,

I followed the torch_geometric tutorial and am using Bipartite Graph for graph classification.

class BipartiteData(Data):
    def __init__(self, x_s, x_t, edge_index):
        super(BipartiteData, self).__init__()
        self.edge_index = edge_index
        self.x_s = x_s
        self.x_t = x_t



    def __inc__(self, key, value):
        if key == 'edge_index':
            return tensor([[self.x_s.size(0)], [self.x_t.size(0)]])
        else:
            return super().__inc__(key, value)

I'm also using Sparse Tensors for the edges and got it to run if I iterate over the Dataloader for batches of size 1, but if I go for bigger batches than that I get:

Traceback (most recent call last):
File "/home/stefan/PycharmProjects/MIPClass/SimMIp/data.py", line 188, in
for i in loader:
File "/home/stefan/anaconda3/envs/pytorch-venv/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 435, in next
data = self._next_data()
File "/home/stefan/anaconda3/envs/pytorch-venv/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 475, in _next_data
data = self._dataset_fetcher.fetch(index) # may raise StopIteration
File "/home/stefan/anaconda3/envs/pytorch-venv/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py", line 47, in fetch
return self.collate_fn(data)
File "/home/stefan/anaconda3/envs/pytorch-venv/lib/python3.8/site-packages/torch_geometric/data/dataloader.py", line 36, in call
return self.collate(batch)
File "/home/stefan/anaconda3/envs/pytorch-venv/lib/python3.8/site-packages/torch_geometric/data/dataloader.py", line 16, in collate
return Batch.from_data_list(batch, self.follow_batch,
File "/home/stefan/anaconda3/envs/pytorch-venv/lib/python3.8/site-packages/torch_geometric/data/batch.py", line 77, in from_data_list
item = item.set_value(value, layout='coo')
File "/home/stefan/anaconda3/envs/pytorch-venv/lib/python3.8/site-packages/torch_sparse/tensor.py", line 171, in set_value
return self.from_storage(self.storage.set_value(value, layout))
File "/home/stefan/anaconda3/envs/pytorch-venv/lib/python3.8/site-packages/torch_sparse/storage.py", line 211, in set_value
assert value.size(0) == self._col.numel()
AssertionError

loader = DataLoader([data, data], batch_size=2)
for i in loader:
     print(net(i))

This is my Data Object: BipartiteData(edge_index=[6598, 5368, nnz=33358], x_s=[5368, 7], x_t=[6598, 4], y=[19])
I suspect that is has to do something with edge_index, so here is the output for data.edge_index as well:

SparseTensor(row=tensor([ 0, 0, 0, ..., 6596, 6597, 6597]),
col=tensor([ 0, 1, 2, ..., 4553, 4526, 4554]),
val=tensor([1., 1., 1., ..., 1., 1., 1.]),
size=(6598, 5368), nnz=33358, density=0.09%)

[rusty1s]

Removing the __inc__ call should omit this error. I'm sorry for the confusion, but there is no need to define __inc__ when using SparseTensor.

[stefansei]

Ah, I see. That part works now. But now the size for the Batched data is not quite working correctly.
If I use a batch size of 2 the resulting size should be (13196, 10736), but it is (6598,10736), i.e, one size is doubled correctly, but the other one didn't change

Batch(batch=[13196], edge_index=[6598, 10736, nnz=66716], ptr=[3], x_s=[10736, 7], x_t=[13196, 4], y=[2])
SparseTensor(row=tensor([ 0, 0, 0, ..., 6597, 6597, 6597]),
col=tensor([ 0, 1, 2, ..., 4554, 9894, 9922]),
val=tensor([1.0000e+00, 1.0000e+00, 1.0000e+00, ..., 1.0000e+00, 6.5990e+03,
6.5990e+03]),
size=(6598, 10736), nnz=66716, density=0.09%)

[rusty1s]

I see. Can you re-install from master and try again?

pip install git+https://github.com/rusty1s/pytorch_geometric.git

This should fix the issue.

[stefansei]

Yes. Worked like a charm. Thanks a lot for your help and this awesome repository 👍