Graclus Issue

[yikang7720]

Hi,

I find a bug exists when I am using the graclus function with weight. This is the code of the graclus network for graph classification. However, I find that if I set the weight of the graclus function as my data weight. The network will run all the time and no results or accuracy provided.

Here is the my Data.
Data(edge_index=[2, 13456], weight=[13456], num_nodes=116, x=[116, 1], y=0).

Here is the code for the graclus network.
`import torch
import torch.nn.functional as F
from torch.nn import Linear

from torch_geometric.data import Batch
from torch_geometric.nn import (
GraphConv,
JumpingKnowledge,
global_mean_pool,
graclus,
max_pool,
)

num_layers = 4
hidden = 16
class Graclus(torch.nn.Module):
def init(self):
super(Graclus, self).init()
self.conv1 = GraphConv(1, 16, aggr='mean')
self.convs = torch.nn.ModuleList()
for i in range(num_layers - 1):
self.convs.append(GraphConv(16, 16, aggr='mean'))
self.jump = JumpingKnowledge(mode='cat')
self.lin1 = Linear(64, 16)
self.lin2 = Linear(16, 1)

def forward(self, data):
    x, edge_index, weight , batch = data.x, data.edge_index, data.weight, data.batch
    x = F.relu(self.conv1(x, edge_index))
    xs = [global_mean_pool(x, batch)]
    for i, conv in enumerate(self.convs):
        x = F.relu(conv(x, edge_index))
        xs += [global_mean_pool(x, batch)]
        if i % 2 == 0 and i < len(self.convs) - 1:
            cluster = graclus(edge_index,num_nodes=x.size(0))
            data = Batch(x=x, edge_index=edge_index, batch=batch)
            data = max_pool(cluster, data)
            x, edge_index, batch = data.x, data.edge_index, data.batch
    x = self.jump(xs)
    x = F.relu(self.lin1(x))
    x = self.lin2(x)
    return x`

[rusty1s]

Can you share your edge_index representation with us? How does data.weight look like? Does it work when enforcing that its values are inside [0, 1)?

[yikang7720]

The edge_index is the connections between different nodes.
It looks like the
'tensor([[ 0, 0, 0, ..., 115, 115, 115],
[ 0, 1, 2, ..., 113, 114, 115]]).'

The edge weight looks like this 'tensor([1.0000, 0.7060, 0.5595, ..., 0.0961, 0.6627, 1.0000])'

I think the value of data.weight is already around [0,1).

[rusty1s]

Can you upload the data somewhere generated via torch.save((edge_index, edge_weight), 'data.pt')?