A julia wrapper for Leiden algorithm (see the paper) to discover and plot the communities of a network.
Leiden algorithm by both CPMVertexPartition (with a resolution parameter γ) and ModularityVertexPartition, as well as Louvain algorithm are implemented.
Technically, the wrapper serves well for huge networks/graphs, weighted or unweighted, directed or undirected, building on SimpleWeightedGraphs.jl. An exponential process over negative node importances or edge weights is recommended during data preparation.
The entire network/graph and the discovered communities can be plotted with the backend GraphPlot.jl. When plotting the whole network, a graph with vertices no more than 5000 is encourgaged. Given a network/graph or a community, the plotting processing will first do a basic partition by ModularityVertexPartition-based Leiden algorithm to ensure 'closer' nodes are rendered with same random colors.
Centrality measures of the nodes and the global cluster coefficients of the graph as well as the local cluster coefficients of the discovered communities could be assessed based on LightGraph.jl.
More detailed documentation will be done soon.
NOTE: the leiden algorithm is implemented by the python package leidenalg, so before doing community discovery, Conda and PyCall have to be installed as follows:
import Pkg
Pkg.add("Conda")
Pkg.add("PyCall")
Pkg.build("PyCall")
using Conda
Conda.pip_interop(true)
#Conda.pip("install", "scipy")
#Conda.pip("install", "numpy")
Conda.pip("install", "leidenalg")
A Demo
include("julia-community.jl")
using DataFrames
import .JuliaCommunity as juliac
nodes = DataFrame(id = [1,2,3,4,5,6,7,8,9,10],
label = ["a","b","c","d","e","f","g","h","i","j"],
importance = [1,5,6,5,5,4,3,3,2,2])
network = DataFrame(from = [1,2,2,2,3,3,4,4,4,5,5,5,6,6,8,8,10,10],
to = [3,3,4,9,2,4,2,3,7,3,6,7,5,8,5,6,7,9],
weight = [1,2,3,1,4,4,2,6,1,1,7,2,5,3,2,4,3,2])
#================================================================
create a JuliaCommunity instance.
task_series is used to name the processing data files (e.g., data/communities-$task_series.csv) and
the svgs (e.g., fig/network-graph-$task_series.svg or fig/community-$(community_id)-$task_series.svg).
================================================================#
jc = juliac.JuliaCommunityInstance(network, nodes = nodes, node_label_field = "label",
node_weighted = true, to_summarise_graph = false, task_series = "demo")
#plot the entire network/graph
juliac.plot_network(jc, line_type="curve", node_size_smoother = 0.8, edge_width_smoother = 1.2)
#run louvain algorithm
#juliac.set_method(jc, jc.methods.louvain)
#run leiden algorithm
#juliac.set_method(jc, jc.methods.CPM) #default
#juliac.set_method(jc, jc.methods.modularity)
#set the CPM γ for leiden algorithm
jc.γ = 0.1
#run community discovering
juliac.discover_communities(jc)
#print the communities discovered
println(jc.communities)
"""
3×2 DataFrame
Row │ c size
│ Int64 Int64
─────┼──────────────
1 │ 1 4
2 │ 2 3
3 │ 3 3
"""
#print the memberships of the communities discovered
println(jc.memberships)
"""
10×2 DataFrame
Row │ id c
│ Int64 Int64
─────┼──────────────
1 │ 1 1
2 │ 2 1
3 │ 3 1
4 │ 4 1
5 │ 5 2
6 │ 6 2
7 │ 7 3
8 │ 8 2
9 │ 9 3
10 │ 10 3
"""
#plot the first community
juliac.plot_community(jc, 1, line_type="curve")
Contributor: Xiaoshan Nian(cen@njust.edu.cn). June, 2021