pytorch_geometric: `return out.scatter_add_(dim, index, src)` RuntimeError: index 765 is out of bounds for dimension 0 with size 765

🐛 Describe the bug

I was doing my graduation project relevant to an NLP task and I encountered this bug when I ran my GNN model training code. This is my first time learning and using torch_geometric, but I have encountered a lot of problems, please help me to solve this problem, I will be grateful！ Simply put, it is a text classification task**(it is a 5 classification task).** In the first step, I have converted the text data into a graph data structure that torch_geometric can identify.

# e.g.
print(dataset[0])
print(dataset[0].num_node_features)
print(dataset[0].num_nodes)
# My word vector dimension is 300, so my num_node_features = 300
> Data(x=[271, 300], edge_index=[2, 1614], y=[1])
> 300
> 271

In the second step, I created my own train_loader with DataLoader, instantiated the GNN model, and defined the device, optimizer, loss function, etc.

dataset.shuffle()
data_size=len(dataset)
train_loader = DataLoader(dataset[:int(data_size * 0.9)], batch_size=3, shuffle=True)

gnn_model = GCN(num_node_feature=300,num_classes=5,hidden_channels=3)
device = torch.device('cuda:0' if torch.cuda.is_available() else "cpu")
gnn_model = gnn_model.to(device)

optimizer = torch.optim.Adam(gnn_model.parameters(), lr=0.005)
criterion = torch.nn.CrossEntropyLoss()

and my GNN model is defined as below.

class GCN(torch.nn.Module):
    def __init__(self, num_node_features, num_classes, hidden_channels):
        super(GCN, self).__init__()
        torch.manual_seed(666)
        self.conv1 = GCNConv(num_node_features, hidden_channels * 2)
        self.conv2 = GCNConv(hidden_channels * 2, hidden_channels)
        self.lin = Linear(hidden_channels, num_classes)

    def forward(self, x, edge_index, batch):

        x = self.conv1(x, edge_index)
        x = x.relu()
        x = self.conv2(x, edge_index)
        x = x.relu()
        x = global_mean_pool(x, batch)  # [batch_size, hidden_channels]

        # 3. Apply a final classifier
        x = F.dropout(x, p=0.2, training=self.training)
        x = self.lin(x)
        return x

In the third step, I wrote the train() function and then tried to train my model. Unfortunately, I encountered the above bug.

def train():
  total_loss= 0
  best_acc = -1
  for d in tqdm(train_loader):
    print(d.x.shape)
    correct = 0
    optimizer.zero_grad()
    out = gnn_model(d.x, d.edge_index, d.batch)
    loss = criterion(out, d.y) 
    loss.backward()
    total_loss+=loss.item()      
    optimizer.step()
    pred = out.argmax(dim=1)  # Use the class with highest probability.
    correct += int((pred == d.y).sum())  # Check against 
    print(f"Train Loss：{loss.item()}")

  total_loss/=len(train_loader)
  acc = correct/len(train_loader)
  return total_loss, acc

for epoch in range(2):
  train()

[<ipython-input-31-50b9ee8e84d4>](https://localhost:8080/#) in train()
      6     correct = 0
      7     optimizer.zero_grad()
----> 8     out = gnn_model(d.x, d.edge_index, d.batch)
      9     loss = criterion(out, d.y)
     10     loss.backward()

[/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py](https://localhost:8080/#) in _call_impl(self, *input, **kwargs)
   1108         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1109                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1110             return forward_call(*input, **kwargs)
   1111         # Do not call functions when jit is used
   1112         full_backward_hooks, non_full_backward_hooks = [], []

[<ipython-input-27-3ada20380b27>](https://localhost:8080/#) in forward(self, x, edge_index, batch)
     22     def forward(self, x, edge_index, batch):
     23 
---> 24         x = self.conv1(x, edge_index)
     25         x = x.relu()
     26         x = self.conv2(x, edge_index)

[/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py](https://localhost:8080/#) in _call_impl(self, *input, **kwargs)
   1108         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1109                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1110             return forward_call(*input, **kwargs)
   1111         # Do not call functions when jit is used
   1112         full_backward_hooks, non_full_backward_hooks = [], []

[/usr/local/lib/python3.7/dist-packages/torch_geometric/nn/conv/gcn_conv.py](https://localhost:8080/#) in forward(self, x, edge_index, edge_weight)
    172                     edge_index, edge_weight = gcn_norm(  # yapf: disable
    173                         edge_index, edge_weight, x.size(self.node_dim),
--> 174                         self.improved, self.add_self_loops)
    175                     if self.cached:
    176                         self._cached_edge_index = (edge_index, edge_weight)

[/usr/local/lib/python3.7/dist-packages/torch_geometric/nn/conv/gcn_conv.py](https://localhost:8080/#) in gcn_norm(edge_index, edge_weight, num_nodes, improved, add_self_loops, dtype)
     62 
     63         row, col = edge_index[0], edge_index[1]
---> 64         deg = scatter_add(edge_weight, col, dim=0, dim_size=num_nodes)
     65         deg_inv_sqrt = deg.pow_(-0.5)
     66         deg_inv_sqrt.masked_fill_(deg_inv_sqrt == float('inf'), 0)

[/usr/local/lib/python3.7/dist-packages/torch_scatter/scatter.py](https://localhost:8080/#) in scatter_add(src, index, dim, out, dim_size)
     27                 out: Optional[torch.Tensor] = None,
     28                 dim_size: Optional[int] = None) -> torch.Tensor:
---> 29     return scatter_sum(src, index, dim, out, dim_size)
     30 
     31 

[/usr/local/lib/python3.7/dist-packages/torch_scatter/scatter.py](https://localhost:8080/#) in scatter_sum(src, index, dim, out, dim_size)
     19             size[dim] = int(index.max()) + 1
     20         out = torch.zeros(size, dtype=src.dtype, device=src.device)
---> 21         return out.scatter_add_(dim, index, src)
     22     else:
     23         return out.scatter_add_(dim, index, src)

RuntimeError: index 765 is out of bounds for dimension 0 with size 765

Environment

• PyG version:(I installed it in Colab)

!pip install torch-cluster -f https://data.pyg.org/whl/torch-1.11.0+cu113.html
!pip install torch-sparse -f https://data.pyg.org/whl/torch-1.11.0+cu113.html
!pip install torch-scatter -f https://data.pyg.org/whl/torch-1.11.0+cu113.html
!pip install torch-geometric

• PyTorch version:1.11.0+cu113

• How you installed PyTorch and PyG (conda, pip, source): pip

• torch-geometric-2.0.4

• torch-scatter-2.0.9

• torch-sparse-0.6.13

• torch-cluster-1.6.0