Perlexities about forward() method on GCN model.

[fedeotto]

Hello everybody,

Thanks to your previous suggestions I have finished to process my graph dataset of inorganic composition storing all entries in a list

>>data_list

[Data(edge_index=[2, 9], x=[3, 112], y=[1]),
 Data(edge_index=[2, 4], x=[2, 112], y=[1]),
 Data(edge_index=[2, 4], x=[2, 112], y=[1]),
...

before, and then in a DataLoader object after

data_loader = DataLoader(data_list, batch_size=100)

Now I have some perplexities to run even simple models on it.
I'm trying with

class Net(torch.nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = GCNConv(112, 64) #here 112 are my node features
        self.conv2 = GCNConv(64, 8)

    def forward(self):
        x = F.relu(self.conv1(data.x, data.edge_index))
        x = self.conv2(x, data.edge_index)
        return x #I want to return just the embeddings (8-dimensional)

So after taking a batch of my train data in DataLoader and instantiating the model,

for data in data_loader:
     break

model = Net()

when I try to run this first batch of data through the model by doing

model(data.x, data.edge_index)

I get

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-334-00cee28ed923> in <module>
----> 1 model(data.x, data.edge_index)

~\anaconda3\lib\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 = [], []

TypeError: forward() takes 1 positional argument but 3 were given

[fedeotto]

Ok I thin I'm figuring out ( a bit slowly) :)

class Net(torch.nn.Module):
    def __init__(self):
        
        super(Net, self).__init__()
        self.conv1 = GCNConv(112, 10)
        self.conv2 = GCNConv(10,8)
        
    def forward(self):
        x = self.conv1(data.x, data.edge_index)
        x = self.conv2(x, edge_index)
        return x

Here the model takes data which is a torch_geometric.data.batch.Batch object and returns the embeddings :) I can also iterate over the data_loader in the following way

for data in data_loader:
    x=model()

This returns the embeddings for the final batch. Any comment /suggestion is highly appreciated :)

[rusty1s]

For iterating via DataLoader, it's best to make forward take input arguments:

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

for data in loader:
    out = model(data.x, data.edge_index)