Hello I am trying to use a GCN layer to perform regression on some data.
My data has 100 nodes and 2 features per node. I tried to interconnect all the nodes to start, but for some reason PyTorch Geometric doesn’t like my edge index matrix.
I am getting the following error:
ValueError: MessagePassing.propagate only supports integer tensors of shape [2, num_messages], torch_sparse.SparseTensor or torch.sparse.Tensor for argument edge_index.
What am I doing wrong?
def make_edge_index(node_size):
rows = []
columns = []
for i in range(node_size):
for j in range(node_size):
rows.append(i)
columns.append(j)
rows = np.asarray(rows)
columns = np.asarray(columns)
index = np.zeros(shape=(2, node_size*node_size))
index[0,:] = rows
index[1,:] = columns
return index
self.input_size = int(node_size/2)
self.output_size = 4
self.edge_matrix = make_edge_index(self.input_size)
self.GCN = geom_nn.conv.GCNConv(2, self.output_size)
self.linear_layer = nn.Linear(node_size, self.input_size)
x = torch.swapaxes(x, 1, 2)
x = x.float()
x = self.linear_layer(x)
# start by scaling down all possible hits using input layer
x = torch.swapaxes(x, 1, 2)
x = self.GCN(x, self.edge_matrix)