Explainer Function Error

[yousha806]

I have been facing this error, the error comes from the module.py file in the under torch.nn.

Please do let me know how to solve this.

[rusty1s]

What is your PyG version? Can you show me the corresponding code snippet?

[yousha806]

This is the PyG version, and below is the code snippet where the error is mentioned in module.py.

[rusty1s]

Thanks! I actually meant a code-snippet to reproduce your issue on my end. Can you share? The error is not very meaningful :(

[yousha806]

Oh! Sorry about that. Below is the code for which it gave me an error.

import numpy as np
import torch

import torch.nn as nn
import torch.nn.functional as F

from torch_geometric.data import Data, DataLoader
from torch_geometric.nn import GATConv, global_mean_pool

import torch_geometric.transforms as T

import matplotlib.pyplot as plt

class GAT(torch.nn.Module):
    def __init__(self):
        super(GAT, self).__init__()
        self.hid = 562
        self.head = 1

        self.conv1 = GATConv(562, self.hid)
        self.conv2 = GATConv(self.hid, self.hid)
        self.lin = nn.Linear(self.hid , 2)

    def forward(self, data):
        x, pos, edge_index = data.Final_Features, data.pos, data.edge_index,

        #x = torch.cat([x, pos], dim=-1)
        x=x.float()
        #print(x.shape)
        x = F.dropout(x, p=0.6, training=self.training)
        x = self.conv1(x.float(), edge_index)
        x = F.elu(x)
        x = F.dropout(x, p=0.6, training=self.training)
        x = self.conv2(x, edge_index)
        x = F.elu(x)
    
        x = global_mean_pool(x, data.batch)
        x = self.lin(x)

        return F.log_softmax(x, dim=1)


device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print('Device', device)

model = GAT().to(device)
from sklearn.metrics import f1_score

optimizer = torch.optim.Adam(model.parameters(), lr=0.005, weight_decay=5e-4)
criterion = torch.nn.CrossEntropyLoss()
model=model.float()
model.train()
for epoch in range(100):
    model.train()
    loss_all = 0
    correct=0
    for batch_data in train_loader:
        batch_data = batch_data.to(device)
        optimizer.zero_grad()
        out = model(batch_data)

        pred = out.argmax(dim=1)  # Use the class with highest probability. 
        correct += int((pred == batch_data.label).sum())  # Check against ground-truth labels.
        loss = criterion(out, batch_data.label)
        loss.backward()
        loss_all += batch_data.num_graphs * loss.item()
        optimizer.step()
    print(f'epoch: {epoch}, loss: {loss_all}')
model.eval()
correct = 0
for data in test_loader:  # Iterate in batches over the training/test dataset.
  out = model(data)  
  #print(out)
  pred = out.argmax(dim=1)  # Use the class with highest probability. 
  correct += int((pred == data.label).sum()) 
  print(f1_score(pred,data.label)) # Check against ground-truth labels.
#print(correct / len(test_loader.dataset))

from torch_geometric.explain import Explainer, GNNExplainer

explainer = Explainer(
    model=model,
    algorithm=GNNExplainer(epochs=100),
          explanation_type='model',
            node_mask_type='attributes',
            edge_mask_type='object',
    model_config=dict(
        mode='binary_classification',
        task_level='graph',
        return_type='raw',  # Model returns log of probability.
    ),
)
g=dataset[0]

explanation = explainer(g.Final_Features, g.edge_index)

This last line with explainer is where the error starts occurring. The dataset array is an array of graphs.
For the dataset, we are batching PyG graphs, where each graph has a format like below:
Data(edge_index=[2, 30], label=0, Final_Features=[6, 562], num_nodes=6)

[rusty1s]

The issue is that you are defining your forward function as forward(self, data) while it needs to be forward(self, x, edge_index, batch). This works fine for me with the applied modifications.