Generating different explanations' subgraphs from the same trained model by using PyG GNNExplainer

[BILLYLIAOWEI]

Hi
I'm using GCN to classify the node on datasets "KarateClub", then I using GNNExplainer to explain node 12. However, when I explain the node twice, GNNExplainer gives me two different subgraph, and them have different node_feat_mask&edge_mask.
I‘m so confused about the different explanations generated from the same trained model.

run code on jupyter notebook
cell one:
`
import torch
from torch_geometric.datasets import KarateClub
import torch.nn.functional as F
from torch_geometric.nn import GCNConv, GNNExplainer
from torch_geometric.datasets import KarateClub
import networkx as nx
import matplotlib.pyplot as plt

dataset = KarateClub()#torch_geometric.datasets

class Net(torch.nn.Module):
def init(self):
super().init()
self.conv1 = GCNConv(dataset.num_node_features, 16)
self.conv2 = GCNConv(16, dataset.num_classes)
pass

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

x = self.conv1(x, edge_index)
x = F.relu(x)
x = F.dropout(x, training=self.training)
x = self.conv2(x, edge_index)

return F.log_softmax(x, dim=1)

pass
pass

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = Net().to(device)
data = dataset[0].to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)
x, edge_index = data.x, data.edge_index

for epoch in range(61):
model.train()
optimizer.zero_grad()
out = model(x, edge_index)
loss = F.nll_loss(out, data.y)
loss.backward()
optimizer.step()

#print('Epoch {} | Loss: {:.4f}'.format(epoch,loss.item()))
model.eval()
_, pred=out.max(dim=1)
#print(pred)
correct = int(pred.eq(data.y).sum().item())
acc = correct / int(data.x.sum())
#print('Accuracy:{:.4f}'.format(acc))
print('Epoch {} | Loss: {:.4f}'.format(epoch,loss.item())+' | Accuracy:{:.4f}'.format(acc))

pass
pass
`

cell two:
explainer = GNNExplainer(model, epochs=60) node_idx = 12 node_feat_mask, edge_mask = explainer.explain_node(node_idx, x, edge_index) ax, G = explainer.visualize_subgraph(node_idx, edge_index, edge_mask, y=data.y) plt.show()

cell three:
explainer = GNNExplainer(model, epochs=61) node_idx = 12 node_feat_mask, edge_mask = explainer.explain_node(node_idx, x, edge_index) ax, G = explainer.visualize_subgraph(node_idx, edge_index, edge_mask, y=data.y,threshold=0.6) plt.show()

[HosseinMousavi]

Hi I'm using GCN to classify the node on datasets "KarateClub", then I using GNNExplainer to explain node 12. However, when I explain the node twice, GNNExplainer gives me two different subgraph, and them have different node_feat_mask&edge_mask. I‘m so confused about the different explanations generated from the same trained model.

run code on jupyter notebook cell one: ` import torch from torch_geometric.datasets import KarateClub import torch.nn.functional as F from torch_geometric.nn import GCNConv, GNNExplainer from torch_geometric.datasets import KarateClub import networkx as nx import matplotlib.pyplot as plt

dataset = KarateClub()#torch_geometric.datasets

class Net(torch.nn.Module): def init(self): super().init() self.conv1 = GCNConv(dataset.num_node_features, 16) self.conv2 = GCNConv(16, dataset.num_classes) pass

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

x = self.conv1(x, edge_index)
x = F.relu(x)
x = F.dropout(x, training=self.training)
x = self.conv2(x, edge_index)

return F.log_softmax(x, dim=1)

pass pass

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') model = Net().to(device) data = dataset[0].to(device) optimizer = torch.optim.Adam(model.parameters(), lr=0.01) x, edge_index = data.x, data.edge_index

for epoch in range(61): model.train() optimizer.zero_grad() out = model(x, edge_index) loss = F.nll_loss(out, data.y) loss.backward() optimizer.step()

#print('Epoch {} | Loss: {:.4f}'.format(epoch,loss.item())) model.eval() _, pred=out.max(dim=1) #print(pred) correct = int(pred.eq(data.y).sum().item()) acc = correct / int(data.x.sum()) #print('Accuracy:{:.4f}'.format(acc)) print('Epoch {} | Loss: {:.4f}'.format(epoch,loss.item())+' | Accuracy:{:.4f}'.format(acc))

pass pass `

cell two: explainer = GNNExplainer(model, epochs=60) node_idx = 12 node_feat_mask, edge_mask = explainer.explain_node(node_idx, x, edge_index) ax, G = explainer.visualize_subgraph(node_idx, edge_index, edge_mask, y=data.y) plt.show()

cell three: explainer = GNNExplainer(model, epochs=61) node_idx = 12 node_feat_mask, edge_mask = explainer.explain_node(node_idx, x, edge_index) ax, G = explainer.visualize_subgraph(node_idx, edge_index, edge_mask, y=data.y,threshold=0.6) plt.show()

Hello BILLYLIAOWEI,
Could you able to run the code and get the result? I want to use PyG and GNNExplainer as well.