dynamic_multiplex (Python)

dynamic_multiplex is a Python package for multiplex community modeling with customizable interlayer ties.

Why this package

Standard multislice approaches (Mucha et al. 2010) connect all layers to all layers, meaning community structure at distant time periods influences assignments everywhere. When applied to temporal data, this creates a pooling problem: the community configuration at 2010 affects the structure detected at 1950. dynamic_multiplex provides explicit control over which layers influence which layers via a layer_links argument (from, to, weight), defaulting to adjacent-only temporal coupling.

Functions

• fit_multilayer_jaccard()
  • Fits Louvain or Leiden communities on each layer.
  • Builds interlayer ties between communities using Jaccard similarity across selected layer pairs.
• fit_multilayer_overlap()
  • Fits Louvain or Leiden communities on each layer.
  • Builds interlayer ties between communities using overlap coefficient across selected layer pairs.
• fit_multilayer_weighted_jaccard()
  • Fits Louvain or Leiden communities on each layer.
  • Builds interlayer ties using node-strength weighted Jaccard similarity across selected layer pairs.
• fit_multilayer_weighted_overlap()
  • Fits Louvain or Leiden communities on each layer.
  • Builds interlayer ties using node-strength weighted overlap coefficient across selected layer pairs.
• fit_multilayer_identity_ties()
  • Fits Louvain or Leiden communities on each layer.
  • Builds interlayer ties only for the same node across selected adjacent layers.
• simulate_and_fit_multilayer()
  • Simulates multiplex layers from a planted partition process and runs one of the three fitting strategies.

Installation (development)

pip install -e ./python_code

Optional algorithms:

pip install -e ./python_code[louvain]
pip install -e ./python_code[leiden]

Quick example

from dynamic_multiplex import simulate_and_fit_multilayer, fit_multilayer_overlap

sim = simulate_and_fit_multilayer(
    directed=True,
    n_nodes=50,
    n_layers=4,
    n_communities=3,
    fit_type="jaccard",
    algorithm="louvain",
    seed=123,
)

print(sim["fit"]["interlayer_ties"].head())

custom_links = [
    {"from": 1, "to": 2, "weight": 1.0},
    {"from": 2, "to": 4, "weight": 0.6},
]

fit_overlap = fit_multilayer_overlap(
    sim["layers"],
    algorithm="leiden",
    layer_links=custom_links,
    min_similarity=0.1,
    add_self_loops=True,
    self_loop_multiplier=1.0,
)

Testing

pip install -e ./python_code[dev,louvain]
pytest