Error with VGAE encoder

🐛 Bug

Error seems to occur with the encoder

data = Data(x=features, edge_index=A)

transform = T.RandomLinkSplit(split_labels=True)
train_data, val_data, test_data = transform(data)

model = VGAE(Encoder(num_features, out_channels))  

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = model.to(device)
x = data.x.to(device)
pos_edge_label_index = train_data.pos_edge_label_index.to(device)

optimizer = torch.optim.Adam(model.parameters(), lr=0.01)

def train():
    model.train()
    optimizer.zero_grad()
    print(x)
    print(train_data.pos_edge_label_index)
    z = model.encode(x, train_data.pos_edge_label_index)
    print(z)
    loss = model.recon_loss(z, train_data.pos_edge_label_index, train_data.neg_edge_label_index)
    
    loss = loss + (1 / data.num_nodes) * model.kl_loss()  
    loss.backward()
    optimizer.step()
    return float(loss)

def test(pos_edge_label_index, neg_edge_label_index):
    model.eval()
    with torch.no_grad():
        z = model.encode(x, train_data.pos_edge_label_index)
        results = model.test(z, pos_edge_label_index, neg_edge_label_index)
    return results[0], results[1], z

Z = []
AUC = []
AP = []
for epoch in range(1, epochs + 1):
    loss = train()
    auc, ap, z = test(test_data.pos_edge_label_index, test_data.neg_edge_label_index)
    Z.append(z)
    AUC.append(auc)
    AP.append(ap)
    print('Epoch: {:03d}, AUC: {:.4f}, AP: {:.4f}'.format(epoch, auc, ap))
    print(z)

Expected behavior

I expect the model to run and return the latent embeddings (z), auc, and ap

I get this error:

RuntimeError                              Traceback (most recent call last)
<ipython-input-48-1ac11cf2e261> in <module>
      3 AP = []
      4 for epoch in range(1, epochs + 1):
----> 5     loss = train()
      6     auc, ap, z = test(test_data.pos_edge_label_index, test_data.neg_edge_label_index)
      7     Z.append(z)

<ipython-input-47-3aaa6b2822ab> in train()
      4     print(x)
      5     print(train_data.pos_edge_label_index)
----> 6     z = model.encode(x, train_data.pos_edge_label_index)
      7     print(z)
      8     loss = model.recon_loss(z, train_data.pos_edge_label_index, train_data.neg_edge_label_index)

~/opt/anaconda3/envs/stGCNG/lib/python3.7/site-packages/torch_geometric/nn/models/autoencoder.py in encode(self, *args, **kwargs)
    154     def encode(self, *args, **kwargs):
    155         """"""
--> 156         self.__mu__, self.__logstd__ = self.encoder(*args, **kwargs)
    157         self.__logstd__ = self.__logstd__.clamp(max=MAX_LOGSTD)
    158         z = self.reparametrize(self.__mu__, self.__logstd__)

~/opt/anaconda3/envs/stGCNG/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-17-d1f128ad6581> in forward(self, x, edge_index)
      7 
      8     def forward(self, x, edge_index):
----> 9         x = self.conv1(x, edge_index).relu()
     10         return self.conv_mu(x, edge_index), self.conv_logvar(x, edge_index)

~/opt/anaconda3/envs/stGCNG/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 = [], []

~/opt/anaconda3/envs/stGCNG/lib/python3.7/site-packages/torch_geometric/nn/conv/gcn_conv.py in forward(self, x, edge_index, edge_weight)
    162                     edge_index, edge_weight = gcn_norm(  # yapf: disable
    163                         edge_index, edge_weight, x.size(self.node_dim),
--> 164                         self.improved, self.add_self_loops)
    165                     if self.cached:
    166                         self._cached_edge_index = (edge_index, edge_weight)

~/opt/anaconda3/envs/stGCNG/lib/python3.7/site-packages/torch_geometric/nn/conv/gcn_conv.py in gcn_norm(edge_index, edge_weight, num_nodes, improved, add_self_loops, dtype)
     60 
     61         row, col = edge_index[0], edge_index[1]
---> 62         deg = scatter_add(edge_weight, col, dim=0, dim_size=num_nodes)
     63         deg_inv_sqrt = deg.pow_(-0.5)
     64         deg_inv_sqrt.masked_fill_(deg_inv_sqrt == float('inf'), 0)

~/opt/anaconda3/envs/stGCNG/lib/python3.7/site-packages/torch_scatter/scatter.py in scatter_add(src, index, dim, out, dim_size)
     27                 out: Optional[torch.Tensor] = None,
     28                 dim_size: Optional[int] = None) -> torch.Tensor:
---> 29     return scatter_sum(src, index, dim, out, dim_size)
     30 
     31 

~/opt/anaconda3/envs/stGCNG/lib/python3.7/site-packages/torch_scatter/scatter.py in scatter_sum(src, index, dim, out, dim_size)
     19             size[dim] = int(index.max()) + 1
     20         out = torch.zeros(size, dtype=src.dtype, device=src.device)
---> 21         return out.scatter_add_(dim, index, src)
     22     else:
     23         return out.scatter_add_(dim, index, src)

RuntimeError: index 1269 is out of bounds for dimension 0 with size 913

The index changes if I restart my kernel (received 1100, 1046, etc.) so I don’t really know where it’s coming from either.

Environment

• OS: 11.5.1
• Python version: 3.7.10
• PyTorch version: 1.9.0
• CUDA/cuDNN version: ?
• GCC version: Apple clang version 12.0.5 (clang-1205.0.22.9)?
• Any other relevant information:

Additional context

I am getting this problem when running the model on my full dataset. Oddly, it works perfectly on a subset of my full data as well as on the CiteSeer dataset, but as soon as I change it to my full data (without modifying anything in the code) it stops working.

The shape of my train_data is: Data(x=[913, 1346], edge_index=[2, 18645], pos_edge_label=[18645], pos_edge_label_index=[2, 18645], neg_edge_label=[18645], neg_edge_label_index=[2, 18645])

[snknitin]

Yes. Thank you. Fortunately, i found my error.

I was building my own InMemoryDataset from .csv files and kept getting the RuntimeError: index 846 is out of bounds for dimension 0 with size 845 when passing the data through any model . There were 846 nodes but since i had to explicitly specify data[node_type].num_nodes = data[node_type].x.size(0) it picked 845. Turns out I didn’t do a reset_index(drop=True) after dropping duplicates in my Node features dataframe, creating an idx mapping for 846 in the (src,dst) edge mapping, and this csv file didn’t have a header, so the first row got discarded when reading the dataframe and shows only 845 nodes

I think the last two assert statements, in your message are missing a ] before the max. Also, shouldn’t it be for edge_type in data.metadata()[1] for HeteroData to get the edge_types?

[rusty1s]

Thanks, I fixed my previous reply.