GATConv with unidirectional bipartite graph

[driss]

please, I created a unidirectional bipartite graph as follows:

G = dgl.heterograph({ ('user', 'rates', 'item') : M.nonzero()})
G.nodes['user'].data['h'] = torch.randn(len(users), 512)
G.nodes['item'].data['h'] = torch.randn(len(items), 512)
G.edges['rates'].data['label'] = weights

I want to apply dgl.nn.GATConv on this graph for the classification of nodes, in this way:

eids = np.arange(G.number_of_edges())
#eids = np.random.permutation(eids)
test_size = int(len(eids) * test_prop)
sampler = dgl.dataloading.MultiLayerFullNeighborSampler(3)
sampler = dgl.dataloading.as_edge_prediction_sampler(sampler)
train_loader = dgl.dataloading.DataLoader(G, eids[2*test_size:], sampler, batch_size=batch_size, shuffle=False, drop_last=False, num_workers=4)
gatconv = GatConv((G.nodes['user'].data['h'].shape[1],G.nodes['item'].data['h'].shape[1), 2,3)
for input_nodes, edge_subgraph, blocks in tqdm(valid_loader):
            edge_subgraph = edge_subgraph.to(torch.device(device))
            blocks = [b.to(torch.device(device)) for b in blocks]
            users = blocks[0].nodes['user'].data['h']
            items = blocks[0].nodes['item'].data['h']
            labels = blocks[0].edata['label']
           res = gatconv(blocks[0], (users, items))

but I get the following error:

AssertionError: Current HeteroNodeDataView has multiple node types, please passing the node type and the corresponding data through a dict.

how i can solve this problem? And thank you in advance.

[mufeili]

Do you have a complete error trace? Also G.nodes['item'].data['h'].shape[1) should be G.nodes['item'].data['h'].shape[1].

[driss]

apparently, I made a mistake at the level of the 1st forward argument of my model. I have to use:
edge_subgraph instead of blocks[0]

[mufeili]

Nice catch! Great to know that.