graph-sd 0.3.1

Mining graphs with Subgroup Discovery

GraphSD

GraphSD (Graph-based Subgroup Discovery) is a Python package for detecting exceptional interaction patterns in graphs. It builds spatio-temporal graphs from position and attribute data, then applies rule-based subgroup discovery and outlier detection techniques to uncover meaningful and rare behaviors.

---

✨ Features

• Directed and multi-directed interaction graph construction
• Subgroup discovery using interpretable rule-based conditions
• Outlier detection and quality-based ranking
• Spatio-temporal interaction filtering using distance and velocity
• Binning and discretization utilities
• Built-in graph visualizations with pattern overlays
• Pure Python: no dependency on Orange3 or external mining engines

---

📦 Installation

Install via PyPI:

pip install graph-sd

---

🚀 Example Usage

from graphsd.mining import DigraphSDMining
from graphsd.utils import make_bins
from graphsd._base import load_data
from graphsd.viz import graph_viz
import networkx as nx

# Load sample position and social data
position_df, social_df = load_data("playground_a")

# Discretize social attributes
social_df = make_bins(social_df)

# Initialize the subgroup discovery engine
dig = DigraphSDMining(random_state=42)

# Build the interaction graph using position and attribute data
dig.read_data(position_df, social_df, time_step=10)

# Discover subgroups with quality constraints
subgroups = dig.subgroup_discovery(
    mode="to",
    min_support=0.2,
    metric="mean",
    quality_measure="global_proportion"
)

# Convert to a DataFrame and print
df = dig.to_dataframe(subgroups)
print(df)

# Visualize the graph and highlighted subgroups
graph_viz(dig.graph, layout=nx.spring_layout)

---

🧠 Code Structure

Module	Purpose
mining.py	Main API for graph construction and subgroup discovery
patterns.py	Logic for rule quality, coverage, and pattern filters
outlier.py	Tools for subgroup scoring and ranking
utils.py	Preprocessing, binning, and distance computations
viz.py	Graph and subgroup visualizations
_base.py	Sample data loader (e.g. load_data("playground_a"))

---

📄 License

This project is licensed under the BSD 3-Clause License.

---

👥 Authors

• Carolina Centeio Jorge – TU Delft
• Cláudio Rebelo de Sá – Leiden University

---

🌐 Links

• 📦 PyPI Package
• 🧑‍💻 GitHub Repository

---

📚 Citation

If you use GraphSD in your research, please cite:

📝 Journal Article (Expert Systems, 2023)

Jorge, C.C., Atzmueller, M., Heravi, B.M., Gibson, J.L., Rossetti, R.J.F., & Rebelo de Sá, C.
"Want to come play with me?" Outlier subgroup discovery on spatio-temporal interactions.
Expert Systems, 40(5), 2023.
https://doi.org/10.1111/exsy.12686

@article{DBLP:journals/es/JorgeAHGRS23,
  author  = {Carolina Centeio Jorge and Martin Atzmueller and Behzad Momahed Heravi and
             Jenny L. Gibson and Rosaldo J. F. Rossetti and Cl{'a}udio Rebelo de S{'a}},
  title   = {"Want to come play with me?" Outlier subgroup discovery on spatio-temporal interactions},
  journal = {Expert Syst. J. Knowl. Eng.},
  volume  = {40},
  number  = {5},
  year    = {2023},
  doi     = {10.1111/EXSY.12686}
}

📘 Conference Paper (EPIA 2019)

Jorge, C.C., Atzmueller, M., Heravi, B.M., Gibson, J.L., Rebelo de Sá, C., & Rossetti, R.J.F.
Mining Exceptional Social Behaviour. In EPIA 2019, LNCS 11805, Springer.
https://doi.org/10.1007/978-3-030-30244-3_38

@inproceedings{DBLP:conf/epia/JorgeAHGSR19,
  author    = {Carolina Centeio Jorge and Martin Atzmueller and Behzad Momahed Heravi and
               Jenny L. Gibson and Cl{'a}udio Rebelo de S{'a} and Rosaldo J. F. Rossetti},
  title     = {Mining Exceptional Social Behaviour},
  booktitle = {Progress in Artificial Intelligence - 19th EPIA 2019},
  series    = {Lecture Notes in Computer Science},
  volume    = {11805},
  pages     = {460--472},
  publisher = {Springer},
  year      = {2019},
  doi       = {10.1007/978-3-030-30244-3_38}
}