Structural Evolution of the Python Package Ecosystem (2016–2025)

🎯 Research Overview

This study presents a comparative topological analysis of the Python Package Index (PyPI) Dependency Network across two temporal snapshots: 2016 and 2025. By modeling the ecosystem as directed graphs, we analyze how the architecture of the network reflects a global transition from web-infrastructure dominance to a data-science and automated testing paradigm.

👥 Authors (FCEN-UBA)

• Carlos Sarraute
• Martina Rosario Pérez
• Juan Ignacio Catania
• Mateo Guerrero Schmidt
• Sofia Gutierrez

Department of Computing, Facultad de Ciencias Exactas y Naturales - Universidad de Buenos Aires

📈 Comparative Network Metrics

Our analysis reveals a massive structural expansion and densification of the ecosystem over a nine-year period:

Metric	2016 Dataset	2025 Dataset
Total Nodes	26,234	295,898
Total Edges	72,252	1,606,337
Average Degree	5.51	10.86
Giant Weakly Connected Component	25,169 nodes	291,204 nodes
Modularity (Louvain)	0.538	0.426

🔍 Key Findings & Industry Evolution

1. The Pivot to Data Science and AI

The comparison of both networks demonstrates a fundamental shift in the ecosystem's "Industry Anchors":

• 2016 Era: Dominated by web development and compatibility tools such as Django, Six, Sphinx, and Distribute.
• 2025 Era: Transition toward Data Science, AI, and Scientific Computing, led by NumPy, Pandas, SciPy, and Torch.

2. Professionalization of Software Work

• Automated Testing: The emergence of Pytest as a top-centrality node in 2025 reflects the universal integration of automated testing and quality assurance into the standard developer workflow.
• Complexity and Density: The increase in average degree from 5.51 to 10.86 indicates that modern software packages now depend on a significantly larger number of libraries, reflecting higher architectural complexity.

3. Structural Dynamics

• Connectivity: While both eras feature a massive Weakly Connected Component, the 2025 network exhibits more "diffuse" community boundaries due to increased interconnection, as evidenced by the drop in modularity.
• Structural Bridges: Libraries such as Requests maintain an influential role across both eras, serving as a primary node for global visibility and communication.

🛠️ Methodology

• Graph Construction: Directed graphs analyzed using NetworkX.
• Centrality Metrics: Evaluation of Degree, Closeness, PageRank, and sampled Betweenness Centrality.
• Community Detection: Implementation of the Louvain Algorithm to detect functional clusters.
• Textual Analysis: Generation of WordClouds from package descriptions to identify community-specific themes.

📚 Bibliography

• [1] Gullikson, Kevin. (2016). Python Dependency Analysis.
• [2] Chugh, V. (2025). Tutorial de pandas en Python: La guía definitiva para principiantes. Datacamp.