Shoudn't the FakeDataset with multiple graphs generate disjoint sets of edges?

[joeda]

Consider

#!/usr/bin/env python3

import torch_geometric.datasets
import torch_geometric.utils
import networkx as nx
import matplotlib.pyplot as plt

dataset = torch_geometric.datasets.FakeDataset(num_graphs=5, avg_num_nodes=5, avg_degree=3, num_channels=5, edge_dim=2, num_classes=10)
G = torch_geometric.utils.to_networkx(dataset.data)
nx.draw(G)
plt.show()

The edge index generated in https://github.com/pyg-team/pytorch_geometric/blob/master/torch_geometric/datasets/fake.py#L84 does not account for multiple graphs

[rusty1s]

Each graph in data is stored individually. If you want to visualize multiple graphs, you can do something like

loader = DataLoader(dataset, batch_size=128)
batch = next(iter(loader)

G = to_networkx(batch)