Error in the Explainer When Specifying Node Index (binary node classification problem)

[Moemenhussein11]

Problem

I am trying to explain the prediction of specific node(s) of my graph. The explainer works very well if I do not specify the index(es); however, it returns an error (see below) if I set the index to index = int.

RuntimeError                              Traceback (most recent call last)
Cell In[46], line 14
      1 explainer = Explainer(
      2     model=model,
      3     algorithm=GNNExplainer(epochs=200),
   (...)
     11     ),
     12 )
     13 node_index = 1
---> 14 explanation = explainer(data.x, data.edge_index, index=node_index)
     15 print(f'Generated explanations in {explanation.available_explanations}')
     17 path = 'feature_importance.png'

File c:\Users\100062576\AppData\Local\Programs\Python\Python312\Lib\site-packages\torch_geometric\explain\explainer.py:205, in Explainer.__call__(self, x, edge_index, target, index, **kwargs)
    202 training = self.model.training
    203 self.model.eval()
--> 205 explanation = self.algorithm(
    206     self.model,
    207     x,
    208     edge_index,
    209     target=target,
    210     index=index,
    211     **kwargs,
    212 )
...
    169 else:
    170     assert False
--> 172 return loss_fn(y_hat.view_as(y), y.float())

RuntimeError: shape '[1]' is invalid for input of size 2

Explainer

I have instantiated the explainer this way:

explainer = Explainer(
    model=model,
    algorithm=GNNExplainer(epochs=200),
    explanation_type='model',
    node_mask_type='attributes',
    edge_mask_type='object',
    model_config=dict(
        mode='binary_classification',
        task_level='node',
        return_type='raw',
    ),
)
node_index = 1
explanation = explainer(data.x, data.edge_index, index=node_index)

explanation.visualize_feature_importance(top_k=10)
explanation.visualize_graph()

My model

from torch_geometric.nn import GATConv, SAGEConv, GCNConv, GraphConv, ChebConv, GatedGraphConv, TAGConv, SSGConv

class GCN(torch.nn.Module):
    def __init__(self, input_dim, hidden_channels, num_layers):
        super().__init__()
        self.conv_layers = torch.nn.ModuleList()
        self.conv_layers.append(GCNConv(input_dim, hidden_channels))

        for _ in range(num_layers - 1):
            self.conv_layers.append(GCNConv(hidden_channels, hidden_channels))

        self.fc = torch.nn.Linear(hidden_channels, 2)

    def forward(self, x, edge_index):
        for conv in self.conv_layers:
            x = conv(x, edge_index).relu()
            x = F.dropout(x, p=0.2, training=self.training)

        x = self.fc(x)
        return x

def train(model, data, optimizer, criterion):
    model.train()
    optimizer.zero_grad()
    out = model(data.x, data.edge_index)
    loss = criterion(out[data.train_mask], data.y[data.train_mask])
    loss.backward()
    optimizer.step()
    return loss.item()

def test(model, data, mask):
    model.eval()
    with torch.no_grad():
        out = model(data.x, data.edge_index)
        pred = out.argmax(dim=1)
    y_true = data.y[mask].cpu().numpy()
    y_pred = pred[mask].cpu().numpy()
    f1 = f1_score(y_true, y_pred)
    return f1

Request for Help

Any insights or solutions to ensure the explainer works when specifying the node index would be greatly appreciated. Thank you!

[Moemenhussein11]

You could set the model_config to mode='multiclass_classification' rather than mode='binary_classification'. This is advisable since the model's output yields two values (from the last layer) instead of just one, which more accurately defines it as a multiclass classification problem.
you can see it at the get_target function in the explainer : if self.model_config.mode == ModelMode.multiclass_classification: return prediction.argmax(dim=-1)