How to use node features which is 2 dimension?

[zzzzzzzzzzzx]

In the pyg data format, x=[num_nodes, num_node_features], in all the explanation and the code I saw, this 'num_node_features' is a one-dimensional vector, that is, the node feature is one-dimensional data, but my node features is a two-dimensional matrix, how can i implement the node features in the pyg data format?

here is an example:
x = torch.tensor([[[2,1,1,3,5,6],[2,1,1,3,5,6]],[[5,6,1,3,5,6],[2,1,1,3,5,6]],[[3,7,1,3,5,6],[2,1,1,3,5,6]],[[12,0,1,3,5,6],[2,1,1,3,5,6]]],dtype=torch.float) y = torch.tensor([0,1,0,1], dtype=torch.float) edge_index = torch.tensor([[0,1,2,1,0,3,3,2], [1,0,1,2,3,0,2,3]],dtype=torch.long) mydata = Data(x=x,y=y,edge_index=edge_index) print(mydata)

The output is Data(edge_index=[2, 5], x=[4, 2, 6], y=[4])

then i use a simple model to train this data:
class Net(torch.nn.Module): def __init__(self): super(Net, self).__init__() self.conv1 = GCNConv(mydata.num_node_features, 2) def forward(self, data): x, edge_index = data.x, data.edge_index x = self.conv1(x, edge_index) return F.log_softmax(x, dim=1)

The whole code is below:
`import torch
import torch.nn.functional as F
from torch_geometric.nn import GCNConv
from torch_geometric.data import Data
from torch_geometric.data import DataLoader

x = torch.tensor([[[2,1,1,3,5,6],[2,1,1,3,5,6]],[[5,6,1,3,5,6],[2,1,1,3,5,6]],[[3,7,1,3,5,6],[2,1,1,3,5,6]],[[12,0,1,3,5,6],[2,1,1,3,5,6]]],dtype=torch.float)
y = torch.tensor([0,1,0,1], dtype=torch.float)
edge_index = torch.tensor([[0,1,2,1,0,3,3,2],
[1,0,1,2,3,0,2,3]],dtype=torch.long)
mydata = Data(x=x,y=y,edge_index=edge_index)
print(mydata)

class Net(torch.nn.Module):
def init(self):
super(Net, self).init()
self.conv1 = GCNConv(mydata.num_node_features, 2)

def forward(self, data):
    x, edge_index = data.x, data.edge_index
    x = self.conv1(x, edge_index)

    return F.log_softmax(x, dim=1)

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = Net().to(device)
data = mydata.to(device)

optimizer = torch.optim.Adam(model.parameters(), lr=0.01, weight_decay=5e-4)

model.train()
for epoch in range(10):
print('epoch',epoch)
optimizer.zero_grad()
out = model(data)
loss = F.nll_loss(out[data.train_mask], data.y[data.train_mask])
loss.backward()
optimizer.step()

model.eval()
_, pred = model(data).max(dim=1)
correct = int(pred[data.test_mask].eq(data.y[data.test_mask]).sum().item())
acc = correct / int(data.test_mask.sum())
print('Accuracy: {:.4f}'.format(acc))`

This code report an error, which is "RuntimeError: size mismatch, m1: [8 x 6], m2: [2 x 2] at /opt/conda/conda-bld/pytorch_1579022034529/work/aten/src/THC/generic/THCTensorMathBlas.cu:290", Obviously this is because of my node feature. SO how can i use the node features which is 2 dimension? THANKS !

[rusty1s]

The MessagePassing API can handle and operate any feature tensor representation via the node_dim argument. It is set to -2 by default, meaning that the second last dimension is used as the node dimension which is used to aggregate information. However, most of the GNNs provided in PyG expect to operate on single feature vector representations.

In your case, I would simply reshape your feature representation to a matrix, since this will result in the computation that I think you are trying to achieve anyway.

[zzzzzzzzzzzx]

Does this sentence "simply reshape your feature representation to a matrix" means reshaping the two-dimensional matrix (node feature) to a 1-dimensional matrix？

[rusty1s]

Yes, something like:

x = ... [num_nodes, F_1, F_2]
x = x.view(num_nodes, F_1 * F_2)

[zzzzzzzzzzzx]

Got it! Thanks !