Process Mining for Python (PM4Py): Bridging the Gap Between Process- and Data Science

Process mining, i.e., a sub-field of data science focusing on the analysis of event data generated during the execution of (business) processes, has seen a tremendous change over the past two decades. Starting off in the early 2000’s, with limited to no tool support, nowadays, several software tools, i.e., both open-source, e.g., ProM and Apromore, and commercial, e.g., Disco, Celonis, ProcessGold, etc., exist. The commercial process mining tools provide limited support for implementing custom algorithms. Moreover, both commercial and open-source process mining tools are often only accessible through a graphical user interface, which hampers their usage in large-scale experimental settings. Initiatives such as RapidProM provide process mining support in the scientific workflow-based data science suite RapidMiner. However, these offer limited to no support for algorithmic customization. In the light of the aforementioned, in this paper, we present a novel process mining library, i.e. Process Mining for Python (PM4Py) that aims to bridge this gap, providing integration with state-of-the-art data science libraries, e.g., pandas, numpy, scipy and scikit-learn. We provide a global overview of the architecture and functionality of PM4Py, accompanied by some representative examples of its usage.

💡 Research Summary

This paper introduces PM4Py (Process Mining for Python), a novel open-source library designed to bridge the gap between the specialized field of process mining and the broader, tool-rich ecosystem of data science. The authors begin by contextualizing the evolution of process mining tools over the past two decades, from early tools with limited support to the current landscape featuring both open-source (e.g., ProM, Apromore) and commercial (e.g., Disco, Celonis) solutions. They identify two critical limitations in existing tools: commercial tools often restrict custom algorithm implementation, and most tools, including open-source ones, are primarily GUI-driven, hindering their use in large-scale, automated experimental settings. While initiatives like RapidProM integrate process mining into scientific workflows, they still offer limited support for algorithmic customization.

To address these gaps, PM4Py is proposed with several core objectives: lowering the barrier for developing and customizing process mining algorithms, enabling seamless integration with state-of-the-art Python data science libraries (pandas, numpy, scikit-learn, etc.), fostering a collaborative ecosystem for sharing code and results, providing comprehensive documentation, and ensuring algorithmic stability through rigorous testing.

The paper details PM4Py’s architecture, which is built on key software engineering principles to maximize understandability, reusability, and scalability for large-scale experiments. A strict separation of concerns is maintained by organizing code into distinct packages for objects (pm4py.objects), algorithms (pm4py.algo), and visualizations (pm4py.visualization). The library heavily utilizes the factory method pattern, providing a single, standardized entry point for each algorithm (e.g., the Alpha Miner) while allowing for different variants and ensuring backward compatibility. This design facilitates easy extension and experimentation.

Functionally, PM4Py supports a wide array of process mining artifacts and techniques. For object management, it handles event logs, event streams, and pandas DataFrames, alongside model representations like Petri nets, process trees, and transition systems, with conversion utilities between them. Its algorithmic portfolio includes mainstream process discovery algorithms (Alpha Miner, Inductive Miner), conformance checking techniques (alignments, token-based replay), model quality metrics (fitness, precision), various filtering methods, case management statistics, and social network analysis. Visualizations leverage established Python libraries such as GraphViz (for Petri nets, process trees), NetworkX, and Pyvis (for interactive social networks).

The authors provide concrete code examples to illustrate PM4Py’s usability. One example demonstrates loading an XES event log, applying the Alpha Miner for process discovery, and visualizing the resulting Petri net. Another example shows performing conformance checking using alignments between a log and a model and printing the detailed alignment results for each trace. These examples showcase the library’s intuitive API and its fit within a standard Python scripting environment.

A section on the maturity of PM4Py highlights its practical adoption and growth trajectory since its first stable release in December 2018. It has been used in university courses, supported academic research projects (e.g., on compliance checking and streaming alignments), and been integrated with other tools like the bupaR library for R. Usage statistics from website visits and PyPI downloads demonstrate growing interest. The library has also obtained official XES standard certification, and its development is supported by a collaborative GitHub repository for issue tracking and contributions.

In conclusion, the paper positions PM4Py as a significant step towards making advanced process mining techniques more accessible, customizable, and integrable within modern data science workflows. By leveraging Python’s extensive ecosystem and a well-designed architecture, PM4Py aims to empower both researchers and practitioners to innovate in the field of process mining.