Large-scale network visualization by k-core decomposition
LaNet-vi draws large networks so that their structure is readable at a glance: nodes are placed in concentric rings by their k-core (or k-dense, or d-core) index, the densest core at the center, with node size following the degree and colors following the index. It is the Python version of the C++ LaNet-vi that produced the well-known Internet AS-level maps (Alvarez-Hamelin, Dall'Asta, Barrat & Vespignani, NIPS 2005; Beiró, Alvarez-Hamelin & Busch, New J. Phys. 2008), and since 5.1.0 it follows the original algorithms.
The Internet at the AS level (CAIDA AS relationships, October 2025: 78,370 ASes,
489,407 links). Left: k-cores 1–149, Tier-1 and hypergiant networks in the red core.
Right: k-denses (m-cores), the triangle-based decomposition.
- 🧅 K-core, k-dense (m-core) and d-core decompositions, weighted k-cores by strength intervals, all as in the C++ LaNet-vi
- 🎯 The LaNet-vi placement: nested components, rings by index, top-core cliques,
and the
pow/logcircle packing of disconnected cores - 🎨 The LaNet-vi look: rainbow or grayscale color scale, gradient edges, seeded edge
sampling, index and degree legends,
--windowzoom, PNG/PDF/SVG output - 🔗 K-connectivity of the shells (
--kconn), the analysis of Beiró, Alvarez-Hamelin & Busch (2008), as in the C++ tool - 📂 Plain edge lists (optionally weighted, directed, compressed) and CAIDA AS-relationship snapshots
- 🐍 CLI and Python API with the same settings, also as a YAML file
- ⚙️ Tested on Python 3.10–3.13; the 78k-node AS graph renders in about a minute
pip install lanet-vi(or uv pip install lanet-vi). Python 3.10 or newer.
lanet-vi visualize --input network.txt --output network.pngimport networkx as nx
from lanet_vi import LaNetConfig, Network
net = Network(nx.karate_club_graph(), LaNetConfig())
net.decompose()
net.visualize("karate.png")The input is an edge list, one source target [weight] per line (# comments,
.gz/.bz2 accepted):
0 1
1 2 2.5
2 0
from lanet_vi import LaNetConfig, Network
from lanet_vi.io.readers import read_caida_snapshot
graph, _ = read_caida_snapshot(
"https://publicdata.caida.org/datasets/as-relationships/serial-1/20251001.as-rel.txt.bz2"
)
net = Network(graph, LaNetConfig())
net.decompose()
net.visualize("internet.png")examples/ has the full scripts behind the pictures above.
--decomp kcores|kdenses|dcores— the decomposition (dcoresneeds--directed)--weighted— the third column is a weight: strength-based cores--edges-percent 0.1— fraction of edges drawn (default 0.5, never below--min-edges)--background white— orblack(default);--color-scheme col|bw|bwi--width 3200 --height 2400— any size and aspect ratio--window 0.25 0.75 0.25 0.75— zoom: the central half of the picture at full size--coord-distribution pow— circle packing of disconnected cores (default:classicrings)--seed 42— reproducible layout and edge sample--detect-communities— Louvain communities colored and outlined on the picture (a 5.x addition)--cores-file cores.csv— also write the decomposition (CSV, or JSON by extension)--config settings.yaml— settings from a file; explicit flags override it
lanet-vi config settings.yaml writes a template with every setting and its default;
lanet-vi info network.txt prints the network statistics; lanet-vi generate makes
random graphs to try things on.
The full documentation is at https://conexdat.github.io/LaNet-vi/ (the same pages
as docs/, with search and an API reference generated from the docstrings):
- Usage guide — every CLI option, the configuration file, the Python API
- Visualization guide — how the picture is built: placement, colors, sizes, edges, legends
- Concepts — k-cores, weighted k-cores, k-denses, k-connectivity, d-cores
- Coming from the C++ LaNet-vi — flag translation and what differs
- Examples — the CAIDA scripts
git clone https://github.com/CoNexDat/LaNet-vi.git
cd LaNet-vi
uv sync --all-extras
uv run pre-commit install
uv run pytestContributions are welcome. main is protected: open a pull request and iterate until CI
and the automatic Copilot review are green. See CONTRIBUTING.md for the
full workflow, coding conventions and release process, and SECURITY.md for
reporting vulnerabilities.
If you use LaNet-vi in your research, please cite the software (GitHub's Cite this
repository button uses CITATION.cff) and the papers behind the method:
-
Alvarez-Hamelin, J.I., Dall'Asta, L., Barrat, A., Vespignani, A. (2006). "Large scale networks fingerprinting and visualization using the k-core decomposition". Advances in Neural Information Processing Systems 18.
-
Beiró, M.G., Alvarez-Hamelin, J.I., Busch, J.R. (2008). "A low complexity visualization tool that helps to perform complex systems analysis". New Journal of Physics.
LaNet-vi 5.x is a from-scratch Python rewrite of the original LaNet-vi (Large Network visualization tool), a C++ program developed since 2005 by Mariano G. Beiró and J. Ignacio Alvarez-Hamelin (Universidad de Buenos Aires / CONICET) together with Alain Barrat, Luca Dall'Asta and Alessandro Vespignani. The C++ tool introduced the concentric k-core layout this package is built on and produced, among others, the Internet AS-level maps that made the method known.
- Original releases (1.x to 3.0.1, last one in January 2016) are published on SourceForge: https://sourceforge.net/projects/lanet-vi/ under the Academic Free License 3.0.
- Original project homepage: http://lanet-vi.fi.uba.ar/.
- The C++ sources are not part of this repository or of the PyPI package; they remain available at the links above.
The Python implementation is released under the MIT License, with the original authors of the C++ version as co-holders of the copyright. The original C++ LaNet-vi remains available under the Academic Free License 3.0 on SourceForge.
- Esteban Carisimo (Python implementation)
- Mariano G. Beiró (original C++ version)
- J. Ignacio Alvarez-Hamelin (original C++ version)