Using ClusterData to split a graph into subgraphs

[yangyangdotcom]

I am trying to use ClustreData to partiton my graphs into 8 subgraphs. However for some of the graph it would return an index error as such:

IndexError: index 236 is out of bounds for dimension 0 with size 235

Any help would be appreciated.

[wsad1]

Hi @yangyangdotcom could you please share the following to help us debug this.

1. Run data.validate and see what happens.
2. Post the full stack trace. it will help us understand the issue better.
3. share the the number of node and edges in you graph and the inputs you passed to ClusterData.

[yangyangdotcom]

Thanks for trplying. Here it is:

<bound method Data.validate of Data(x=[219, 20], edge_index=[2, 2422], edge_attr=[2422], seq='LDKQQVQLLAEMCILIDENDNKIGAETKKNCHLNENIEKGLLHRAFSVFLFNTENKLLLQQRSDAKITFPGCFTNTCCSHPLSNPAELEESDALGVRRAAQRRLKAELGIPLEEVPPEEINYLTRIHYKAQSDGIWGEHEIDYILLVRKNVTLNPDPNEIKSYCYVSKEELKELLKKAASGEIKITPWFKIIAATFLFKWWDNLNHLNQFVDHEKIYRM', name=/Users/usr/Documents/Classification/Graph/Data/raw/2ICJ.pdb, adj=[219, 219])>

Traceback (most recent call last):
File "/Users/usr/Documents/Classification/Graph-subgraph/testing.py", line 63, in
partiton_graph()
File "/Users/usr/Documents/Classification/Graph-subgraph/testing.py", line 22, in partiton_graph
cluster = ClusterData(data[0],4,save_dir = "/Users/usr/Documents/Classification/Graph-subgraph/subgraph_data/")
File "/Users/usr/opt/anaconda3/envs/PPI_GNN_seqvec/lib/python3.9/site-packages/torch_geometric/loader/cluster.py", line 58, in init
self.data = self.permute_data(data, perm, adj)
File "/Users/usr/opt/anaconda3/envs/PPI_GNN_seqvec/lib/python3.9/site-packages/torch_geometric/loader/cluster.py", line 66, in permute_data
out[key] = value[node_idx]
IndexError: index 219 is out of bounds for dimension 0 with size 219

My graph has 219 nodes and 1211 edges. I pass in the data from the first bullet point to the ClusterData

[wsad1]

Sorry i meant call data.validate(). From the error it looks like your edge_index might have values greater thank 218, but the nodes are indexed from 0-218, can you confirm this.

[yangyangdotcom]

It returns True when I call data.validate(). That is correct, my graph has more edges than the nodes. I try to partition other dataset that has more edges than nodes and it works. Not sure why it doesn't work on my dataset.

[wsad1]

Its ok for your graph to have more edges than nodes. But the edge should be between nodes that exist, so if your graph has 219 nodes then edge_index has to have entries between [0,218].

[yangyangdotcom]

I think I found the problem. This happen when a path is given as the save_dir. Not sure why, but the error disappear after I remove the save_dir.