GitHub - nveldt/HyperModularity.jl (original) (raw)

HyperModularity

This package contains code for hypergraph modularity clustering algorithms, based on the generative model described in

Generative hypergraph clustering: from blockmodels to modularity
Philip S. Chodrow, Nate Veldt, Austin R. Benson
preprint

Package Installation

To install package

using Pkg
Pkg.add("HyperModularity")
using HyperModularity

Data available

For a full list of all datasets:

hypermodularity_datasets()

Loading data example:

dataset = "contact-high-school-classes"
maxsize = 5	# max hyperedge size
minsize = 2	# min hyperedge size
return_labels = true
H, L = read_hypergraph_data(dataset,maxsize,minsize,return_labels)

Some Examples

Run graph Louvain algorithm on clique expanded graph

gamma = 1.0 # default resolution parameter
Z_g = CliqueExpansionModularity(H,gamma) # see code for default parameters

# Compute MLE resolution parameter given clustering Z_g
(ωᵢ, ωₒ) = computeDyadicResolutionParameter(H, Z_g; mode = 0)
γ_mle = (ωᵢ - ωₒ)/(log(ωᵢ) - log(ωₒ))
loglike = dyadicLogLikelihood(H, Z, ωᵢ, ωₒ)

Run all-or-nothing Louvain algorithm

n = size(H,2)
Z_ = collect(1:n) # trivial clustering

# all or nothing aggregator: p -> [length(p) == 1, sum(p)]
# This gives a starter estimate for Ω, from a trivial clustering Z_
Ω = estimateΩEmpirically(H, Z_; aggregator = p -> [length(p) == 1, sum(p)])

Z = AON_Louvain(H,Ω)

# Alternatively, one can learn Ω from graph Louvain solution Z_g
Ω = estimateΩEmpirically(H, Z_g; aggregator = p -> [length(p) == 1, sum(p)])
Z = AON_Louvain(H,Ω)

Additional examples

See demo notebooks in demos folder for other examples of how to use the code.