tau-community-detection 1.2.0

Community detection via Louvain/Leiden + Genetic Algorithm

TAU Community Detection

tau-community-detection implements TAU, an evolutionary community detection algorithm that couples genetic search with Leiden refinements. It is designed for scalable graph clustering with configurable hyper-parameters and multiprocessing support.

---

Highlights

• Evolutionary search: Maintains a population of candidate partitions and applies crossover/mutation tailored for graph clustering.
• Leiden optimization: Refines every candidate with Leiden to ensure modularity gains.
• Multiprocessing aware: Utilises worker pools for population optimization.
• Deterministic options: Accepts a user-specified random seed for reproducibility.
• Simple API: Access everything through the TauClustering class.

---

Installation

The project targets Python 3.10 or newer.

pip install tau-community-detection

To work from a clone, install the package in editable mode inside a virtual environment:

git clone https://github.com/HillelCharbit/community_TAU.git
cd community_TAU
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
pip install -e .

---

Quick Start (Python API)

from tau_community_detection import TauClustering
import networkx as nx

graph = nx.read_adjlist("path/to/graph.adjlist")

clustering = TauClustering(
    graph,
    population_size=80,
    max_generations=250,
)
membership, modularity_history = clustering.run()

print("community for node 0:", membership[0])
print("best modularity:", modularity_history[-1])

Need detailed per-generation metrics? Call run(track_stats=True) to receive (membership, modularity_history, generation_stats) where generation_stats is a list of dictionaries containing time and fitness diagnostics.

---

Graph input

To optimize for very large graphs or when using many worker processes, it is recommended to pass a file path (e.g., to an .adjlist or edge list file) directly to TauClustering rather than a pre-loaded graph object. This allows efficient memory sharing.

Supported input:

• File path to a graph in common NetworkX or igraph format (auto-detects weighting and structure).
• Already-loaded networkx.Graph or igraph.Graph objects.

By default, the loader auto-detects whether the graph is weighted based on the file or graph structure. You can override this by setting is_weighted=True or False when constructing TauClustering or its config object; if your override disagrees with the detected type, a warning is issued and auto-detection is used.

Examples:

from tau_community_detection import TauClustering

# Recommended for large graphs:
clustering = TauClustering("mygraph.graph", population_size=40, max_generations=30)

# In-memory NetworkX graph:
import networkx as nx
g = nx.read_adjlist("mygraph.adjlist")
clustering2 = TauClustering(g, population_size=40, max_generations=30)

# Force unweighted input (ignore/strip weights if present):
clustering3 = TauClustering("mygraph.graph", population_size=40, max_generations=30, is_weighted=False)

For details on accepted formats, see below.

---

Configuration

All algorithm hyper-parameters live on the TauConfig dataclass. You can pass a custom configuration instance to TauClustering or adjust attributes on the default one. Key fields include:

• population_size: number of partitions maintained per generation.
• max_generations: upper bound on evolutionary iterations.
• elite_fraction / immigrant_fraction: govern selection pressure.
• stopping_generations / stopping_jaccard: convergence checks based on membership stability.
• random_seed: makes runs reproducible across processes.

See src/tau_community_detection/config.py for the complete list.

---

Development

pip install -r requirements-dev.txt
make lint
make test

To build local distributions:

make build

Continuous Integration

• GitHub Actions run lint, tests, and package builds on pushes and pull requests.
• Set the CODECOV_TOKEN secret to upload coverage reports.

Publishing

1. Bump the version in setup.cfg/pyproject.toml and commit.
2. Tag the release with git tag vX.Y.Z && git push --tags.
3. Run the Publish Package workflow (defaults to TestPyPI). For PyPI, supply the pypi input and ensure PYPI_API_TOKEN is set. Use TEST_PYPI_API_TOKEN for dry runs.

---

License

Released under the MIT License. See LICENSE for details.