RuntimeError: view size is not compatible with input tensor's size in TSAGCN

[Borzyszkowski]

When using TSAGCN from torch_geometric_temporal, I encountered the following error:

File "lib/python3.8/site-packages/torch_geometric_temporal/nn/attention/tsagcn.py", line 260, in forward
    y = self._adaptive_forward(x, y)
  File "lib/python3.8/site-packages/torch_geometric_temporal/nn/attention/tsagcn.py", line 237, in _adaptive_forward
    A2 = self.conv_b[i](x).view(N, self.inter_c * T, V)
RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.

This error occurs in _adaptive_forward() when calling .view() on a tensor that is not memory-contiguous.

Expected Behavior
The model should process the input without error.

Actual Behavior
The model crashes due to .view() being used on a non-contiguous tensor.

System Information

torch==2.4.1+cu124
torch-geometric==2.6.1
torch-geometric-temporal==0.54.0
torch_cluster==1.6.3+pt24cu124
torch_scatter==2.1.2+pt24cu124
torch_sparse==0.6.18+pt24cu124
torch_spline_conv==1.2.2+pt24cu124
torchaudio==2.4.1+cu124
torchgeometry==0.1.2
torchvision==0.19.1+cu124
Python version: 3.8.10
OS: ubuntu20

Example to reproduce:

import torch
from torch_geometric.data import Data
from torch_geometric_temporal.nn.attention import AAGCN

# Define the graph that produces the error
x = torch.rand(2, 80, 1556, 1)
edge_index = torch.randint(0, 2, (2, 18456), dtype=torch.long)
y = torch.rand(2, 2)  

# Create the graph object
graph = Data(x=x, edge_index=edge_index, y=y)

# Create the AAGCN layer (Attention Aggregated Graph Convolutional Network)
aagcn = AAGCN(in_channels=1,
              out_channels=16,
              edge_index=edge_index,
              num_nodes=2,
              adaptive=True,
              attention=True)

# Forward pass (this is where the original issue happens)
x = x.permute(0, 3, 1, 2).contiguous()  # Batch, Features_In, Temporal_In, Node_Number

 # This will raise the 'view' error in some cases
output = aagcn(x) 

current code in torch_geometric_temporal/nn/attention/tsagcn.py#237
A2 = self.conv_b[i](x).view(N, self.inter_c * T, V)

proposed fix:
A2 = self.conv_b[i](x).reshape(N, self.inter_c * T, V)