RuntimeError: expected scalar type Double but found Float

[siserte]

Hello,
I had working my GNN like this:

class GCN(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = GCNConv(3, 16)
        self.conv2 = GCNConv(16, 32)
        self.conv3 = GCNConv(32, 16)
        self.fc1 = Linear(16, 10)
       
    def forward(self, data):
        x, edge_index, edge_attr = data.x, data.edge_index, data.edge_attr
        x = F.relu(self.conv1(x, edge_index))
        x = F.relu(self.conv2(x, edge_index))
        x = F.relu(self.conv3(x, edge_index))
        x = self.fc1(x)
        return x  

So far, so good. But now I've decided to add attributes to the edges updating my layers like this:

        x = F.relu(self.conv1(x, edge_index, edge_attr))
        x = F.relu(self.conv2(x, edge_index, edge_attr))
        x = F.relu(self.conv3(x, edge_index, edge_attr))
```

And I get the **RuntimeError: expected scalar type Double but found Float**

This is the execution output with information about the model and the error log:

```
DataParallel(
  (module): GCN(
    (conv1): GCNConv(3, 16)
    (conv2): GCNConv(16, 32)
    (conv3): GCNConv(32, 16)
    (fc1): Linear(in_features=16, out_features=10, bias=True)
  )
)
Model parameters: 1306
DataLoader lenght: 704
8 Data(x=[382356, 3], edge_index=[2, 75516], edge_attr=[75516, 1], y=[382356, 10])
# Graphs: 4
# Features: 3
# Are graphs undirected? True
---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-45-949585f7a635> in <module>
     38         break
     39 
---> 40     model, train_loss, valid_loss = train_model(model, n_batch, patience, EPOCHS)
     41 
     42     # visualize the loss as the network trained

<ipython-input-43-60213e6cffec> in train_model(model, batch_size, patience, n_epochs)
     88             #print(len(data_list))
     89             optimizer.zero_grad()
---> 90             output = model(data_list)
     91             #print(f'Outside model - num graphs: {output.size(0)}')
     92             y = torch.cat([data.y for data in data_list]).to(output.device)

/apps/TORCH/1.9.0a0/lib/python3.7/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1049         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1050                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1051             return forward_call(*input, **kwargs)
   1052         # Do not call functions when jit is used
   1053         full_backward_hooks, non_full_backward_hooks = [], []

/apps/PYTHON/3.7.4_ML/lib/python3.7/site-packages/torch_geometric/nn/data_parallel.py in forward(self, data_list)
     56                 data_list, follow_batch=self.follow_batch,
     57                 exclude_keys=self.exclude_keys).to(self.src_device)
---> 58             return self.module(data)
     59 
     60         for t in chain(self.module.parameters(), self.module.buffers()):

/apps/TORCH/1.9.0a0/lib/python3.7/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1049         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1050                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1051             return forward_call(*input, **kwargs)
   1052         # Do not call functions when jit is used
   1053         full_backward_hooks, non_full_backward_hooks = [], []

<ipython-input-43-60213e6cffec> in forward(self, data)
     59         """
     60         x = F.relu(self.conv1(x, edge_index, edge_attr))
---> 61         x = F.relu(self.conv2(x, edge_index, edge_attr))
     62         x = F.relu(self.conv3(x, edge_index, edge_attr))
     63 

/apps/TORCH/1.9.0a0/lib/python3.7/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1049         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1050                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1051             return forward_call(*input, **kwargs)
   1052         # Do not call functions when jit is used
   1053         full_backward_hooks, non_full_backward_hooks = [], []

/apps/PYTHON/3.7.4_ML/lib/python3.7/site-packages/torch_geometric/nn/conv/gcn_conv.py in forward(self, x, edge_index, edge_weight)
    180                     edge_index = cache
    181 
--> 182         x = self.lin(x)
    183 
    184         # propagate_type: (x: Tensor, edge_weight: OptTensor)

/apps/TORCH/1.9.0a0/lib/python3.7/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1049         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1050                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1051             return forward_call(*input, **kwargs)
   1052         # Do not call functions when jit is used
   1053         full_backward_hooks, non_full_backward_hooks = [], []

/apps/PYTHON/3.7.4_ML/lib/python3.7/site-packages/torch_geometric/nn/dense/linear.py in forward(self, x)
    107     def forward(self, x: Tensor) -> Tensor:
    108         """"""
--> 109         return F.linear(x, self.weight, self.bias)
    110 
    111     @torch.no_grad()

/apps/TORCH/1.9.0a0/lib/python3.7/site-packages/torch/nn/functional.py in linear(input, weight, bias)
   1845     if has_torch_function_variadic(input, weight):
   1846         return handle_torch_function(linear, (input, weight), input, weight, bias=bias)
-> 1847     return torch._C._nn.linear(input, weight, bias)
   1848 
   1849 

RuntimeError: expected scalar type Double but found Float
```

I've just in case, I've also tried with layers GATConv and GATv2Conv (I've seen that they are compatible with attr_edge [here](https://pytorch-geometric.readthedocs.io/en/latest/notes/cheatsheet.html), but it either does not work. Although for these layers the error only says: **AssertionError: **.

Thanks for your help.

[rusty1s]

You should confirm that both data.x.dtype and data.edge_attr.dtype are of dtype == torch.float.

[siserte]

Thanks @rusty1s ,
I've solved it converting the data and the model:

x = torch.tensor(dat)
x = x.double()
y = torch.tensor(target)
y = y.double()
graph = Data(x=x, edge_index=edges, y=y, edge_attr=attrs)

model.double()

[rusty1s]

Super. I suggest to move everything to float. Usually, the higher precision is not needed in GNNs.