asbm 0.1.4.dev5

Find group structures in networks

asbm

Infer network groups and properties with assortative stochastic block models

Maximilian Jerdee

This Python package uses Bayesian inference to find meaningful groupings of nodes in networks.

We implement a general assortative SBM that unifies the standard SBM and the planted partition model under a common framework. Its parameters directly measure the violation of each model's assumptions: the assortative preference ρ_in/ρ_out captures departure from in/out symmetry, while per-group variation coefficients v_in and v_out measure heterogeneity across groups. The standard SBM, planted partition model, and the Zhang–Peixoto hybrid emerge as special cases, enabling exact Bayesian model comparison between them.

For each model, the package includes algorithms to:

• Find consensus estimates of the group structure
• Infer global network parameters (assortativity, group sizes)
• Score held-out edges via posterior predictive likelihood

Installation

Implementations are available for Python, R, and Julia.

Python

pip install asbm

Or build locally from the repository root:

pip install .

R

install.packages("asbm", repos = "https://maxjerdee.r-universe.dev")

Julia

using Pkg
Pkg.add(url="https://github.com/maxjerdee/asbm", subdir="bindings/julia")

Building from source requires CMake and a C++17 compiler. Run Pkg.build("ASBM") after installation to compile the native library.

Quickstart

Python

import asbm
import networkx as nx

G = nx.read_gml("examples/data/dolphins.gml", label="id")

config = asbm.Config(
    model="general_asbm",
    degree_correction=True,
)

result = asbm.fit(config, G)

print(result.mdl_partition)
print(result.mdl_value)
print(result.consensus_partition())

To score held-out edges:

G_train = nx.read_gml("examples/data/train.gml")
G_test  = nx.read_gml("examples/data/test.gml")

result = asbm.fit(asbm.Config(model="general_asbm"), G_train)
score  = result.log_posterior_predictive(G_test)

R

library(asbm)
library(igraph)

G <- read_graph("examples/data/dolphins.gml", format = "gml")

result <- fit(G,
              model             = "general_asbm",
              degree_correction = TRUE,
              num_chains        = 4,
              samples_per_chain = 100,
              seed              = 42)

print(result$mdl_value)
print(result$mdl_partition)
print(result$consensus_partition)

Julia

using ASBM, Graphs

g = cycle_graph(62)  # or load via GraphIO

result = fit(g;
    model             = "general_asbm",
    degree_correction = true,
    num_chains        = 4,
    samples_per_chain = 100,
    seed              = 42)

println(result.mdl_value)
println(result.mdl_partition)
println(result.consensus_partition)

For the full API, including posterior predictive evaluation and the samples schema, see the package documentation.