multiclass_classification graph explainer

[JFetish]

I'm trying to use the explainer module for multiclass graph classification problems, but I'm stuck with some errors.

Traceback (most recent call last):
  File "/home/dldx/Desktop/CRK/classification/tt.py", line 129, in <module>
    loss = explainer.algorithm.train(
  File "/home/dldx/anaconda3/envs/crk/lib/python3.9/site-packages/torch_geometric/explain/algorithm/pg_explainer.py", line 123, in train
    z = get_embeddings(model, x, edge_index, **kwargs)[-1]
  File "/home/dldx/anaconda3/envs/crk/lib/python3.9/site-packages/torch_geometric/utils/embedding.py", line 48, in get_embeddings
    model(*args, **kwargs)
  File "/home/dldx/anaconda3/envs/crk/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
TypeError: forward() takes 2 positional arguments but 3 were given

code

class Cheb(nn.Module):
    def __init__(self, feature_dim, hidden_dim=32, output_dim=256, dropout=0.4, use_residue=False):
        super(Cheb, self).__init__()
        self.dropout = dropout
        self.hidden = hidden_dim
        self.cconv1 = ChebConv(feature_dim, hidden_dim, aggr='sum', K=3)
        self.cconv2 = ChebConv(hidden_dim, hidden_dim * 2, aggr='sum', K=3)
        self.cconv3 = ChebConv(hidden_dim * 2, hidden_dim * 4, aggr='sum', K=3)

        self.relu = nn.LeakyReLU()
        self.dropout = nn.Dropout(dropout)
        self.classification = nn.Sequential(
            nn.Linear(self.hidden * 4, 1024),
            nn.ReLU(),
            nn.Linear(1024, 512),
            nn.ReLU(),
            nn.Linear(512, 4)
        )
    def forward(self, data):
        # 获取小分子和蛋白质输入的结构信息
        protein_feature, protein_index, protein_batch = data.x, data.edge_index, data.batch
        # 对小分子进行卷积操作
        protein_feature = self.cconv1(protein_feature, protein_index)
        protein_feature = self.relu(protein_feature)
        # protein_feature = self.bn1(protein_feature)

        protein_feature = self.cconv2(protein_feature, protein_index)
        protein_feature = self.relu(protein_feature)
        # protein_feature = self.bn2(protein_feature)

        protein_feature = self.cconv3(protein_feature, protein_index)
        # protein_feature = self.relu(protein_feature)
        # protein_feature = self.bn3(protein_feature)

        # 对卷积后的小分子进行图的最大值池化
        # print(protein_feature.view(16, -1, self.hidden * 4))
        protein_feature = gmp(protein_feature, protein_batch)

        protein_feature = self.classification(protein_feature)

        return protein_feature

loader = DataLoader(test_data, batch_size=1, shuffle=False)
    for batch in loader:
        print(batch.x)
    explainer = Explainer(model=model, algorithm=PGExplainer(epochs=30, lr=0.003),
        explanation_type='phenomenon',
        edge_mask_type='object',
        model_config=dict(mode='multiclass_classification', task_level='graph', return_type='raw', ),
        threshold_config=dict(threshold_type='topk', value=10)
    )
for epoch in range(20):
     for batch in loader:
     loss = explainer.algorithm.train(epoch, model, batch.x, batch.edge_index, target=batch.y)

data

DataBatch(x=[78, 44], edge_index=[2, 2714], y=[1], id=[1], size=[1], batch=[78], ptr=[2])
DataBatch(x=[86, 44], edge_index=[2, 3052], y=[1], id=[1], size=[1], batch=[86], ptr=[2])
DataBatch(x=[70, 44], edge_index=[2, 2302], y=[1], id=[1], size=[1], batch=[70], ptr=[2])
DataBatch(x=[87, 44], edge_index=[2, 3252], y=[1], id=[1], size=[1], batch=[87], ptr=[2])
DataBatch(x=[67, 44], edge_index=[2, 1378], y=[1], id=[1], size=[1], batch=[67], ptr=[2])
DataBatch(x=[73, 44], edge_index=[2, 1598], y=[1], id=[1], size=[1], batch=[73], ptr=[2])
DataBatch(x=[86, 44], edge_index=[2, 3116], y=[1], id=[1], size=[1], batch=[86], ptr=[2])

[rusty1s]

You should modify your model to only take in the actual tensors as input, e.g.:

def forward(self, x, edge_index, batch)

These parameters will get passed to the model in

explainer.algorithm.train(epoch, model, batch.x, batch.edge_index, batch=batch.batch, target=batch.y)

[JFetish]

Thank you!