Node2Vec - Segmentation Fault

🐛 Bug

When I run Node2Vec on a graph, I receive a Segmentation fault (core dumped) message.

To Reproduce

Steps to reproduce the behavior:

1. Edge List

I will post this in the next comment (the file is long)

2. Script

import networkx as nx
import torch
from torch_geometric.nn import Node2Vec
import numpy as np


walk_length = 40

walks_per_node = 10

context_size = 5

p = 1
q = 1

def txt2nx(file):

    graph = nx.read_edgelist(file)

    return graph


if __name__ == "__main__":

    def train():
        model.train()
        total_loss = 0
        for pos_rw, neg_rw in loader:
            optimizer.zero_grad()
            loss = model.loss(pos_rw.to(device), neg_rw.to(device))
            loss.backward()
            optimizer.step()
            total_loss += loss.item()
        return total_loss / len(loader)

    def train_epoch():
        print()

        for _ in range(5):
            loss = train()
            print(loss)

    G = txt2nx('./test.edge_list')

    G = nx.relabel.convert_node_labels_to_integers(G)

    edge_index = torch.tensor(
                            np.array(G.edges()),
                            dtype=torch.long
                            ).t().contiguous()

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

    model = Node2Vec(
        edge_index,
        embedding_dim=8,
        walk_length=walk_length,
        context_size=context_size,
        walks_per_node=walks_per_node,
        num_negative_samples=1,
        p=p,
        q=q,
        sparse=True,
    ).to(device)

    loader = model.loader(batch_size=128, shuffle=True)  # , num_workers=4)
    optimizer = torch.optim.SparseAdam(model.parameters(), lr=0.01)

    # Train until delta loss has been reached
    train_epoch()

3. Output from running script

Segmentation fault (core dumped)

Expected behavior

I would have expected no errors training the embeddings with Node2Vec.

Environment

• OS: Ubuntu 20.04
• Python version: 3.8.2
• PyTorch version: 1.5.1
• CUDA/cuDNN version: (Ran this on CPU, but have 10.2)
• GCC version: gcc (Ubuntu 9.3.0-10ubuntu2) 9.3.0
• Any other relevant information:

Additional context

I tested it with networkx isolates (nx.algorithms.isolate.isolates(G)) and there were no isolated nodes.

[davecerr]

I had a similar error. I found that not using networkx resolved the problem. Maybe you can save your edge list in a text file and read it from there instead?

[dansontong]

Thanks @davecerr. I solved it by saving edge list in a text file and then read it again, just like what @davecerr says.