Add assortativity to torch_geometric.utils

[EdisonLeeeee]

This PR implements the computation of degree assortativity coefficient, as described in Mixing patterns in networks. Assortativity is a preference for a network's nodes to attach to others that are similar in some way. A larger assortativity indicates nodes tend to be connected with other nodes with similar degree values, and vice versa. Link to Slack Discussion

The implementation is based on networkx.degree_assortativity_coefficient but much faster than using networkx.

Test

• Generate a graph (without duplicated edges)

import torch
from torch_geometric.utils import coalesce

torch.manual_seed(12345)
num_nodes = 10000
num_edges = 1000000
edge_index = torch.randint(0, num_nodes, size=(2, num_edges))
edge_index = coalesce(edge_index)

• networkx.degree_assortativity_coefficient

import networkx as nx
nxg = nx.from_edgelist(edge_index.t().tolist(), create_using=nx.DiGraph)
t = time.perf_counter()
result = nx.degree_assortativity_coefficient(nxg)
print(f"result={result}, time={time.perf_counter()-t}")

>>> result=-0.0007873635931715391, time=0.48579016700000466

• torch_geometric.utils.assortativity (On CPU)

from torch_geometric.utils import assortativity
t = time.perf_counter()
result = assortativity(edge_index)
print(f"result={result}, time={time.perf_counter()-t}")

>>> result=-0.0007878749747760594, time=0.07563883299985719

• torch_geometric.utils.assortativity (On GPU)

>>> result=-0.0007857615128159523, time=0.0036659350153058767

The slight difference in results is caused by two different frameworks (Numpy and PyTorch).

[rusty1s]

Looks good to me. Please also add it to the CHANGELOG.md.

[codecov]

Codecov Report

Merging #5587 (9e56273) into master (6ca2332) will increase coverage by 0.02%.
The diff coverage is 96.77%.

@@            Coverage Diff             @@
##           master    #5587      +/-   ##
==========================================
+ Coverage   83.85%   83.88%   +0.02%     
==========================================
  Files         348      349       +1     
  Lines       19169    19200      +31     
==========================================
+ Hits        16074    16105      +31     
  Misses       3095     3095              

Impacted Files	Coverage Δ
torch_geometric/utils/assortativity.py	96.77% <96.77%> (ø)
torch_geometric/utils/coalesce.py	100.00% <0.00%> (+3.12%)	⬆️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more