An implementation of a process hypothesis testing technique using the Earth Mover's Distance and a permutation test.

• This is the process hypothesis testing technique proposed in my Master's Thesis "Hypothesis Testing for Processes: Bridging Statistical Methods and Treatment Disparities in Healthcare"
  • The thesis is based on commit 9aaac79 (v0.1.0)
  • The experiments conducted in the thesis are in the cpitsch/pcomp-experiments repository

from pm4py import read_xes

from pcomp.emd.comparators.permutation_test import (
    Timed_Levenshtein_PermutationComparator,
)

log_1 = read_xes("path/to/log_1.xes")
log_2 = read_xes("path/to/log_2.xes")
comparator = Timed_Levenshtein_PermutationComparator(
    log_1, log_2, distribution_size=10_000, seed=1337, weighted_time_cost=True
)
result = comparator.compare()
print(f"P-Value: {result.pvalue}")
result.plot().show()

The project also contains an implementation of the P-P-UP (Process-Process-Unknown Process) test proposed in "Statistical Tests and Association Measures for Business Processes" by Leemans et al.:

from pm4py import read_xes

from pcomp.emd.comparators.bootstrap import ControlFlowBootstrapComparator

log_1 = read_xes("path/to/log_1.xes")
log_2 = read_xes("path/to/log_2.xes")

comparator = ControlFlowBootstrapComparator(
    log_1,
    log_2,
    bootstrapping_dist_size=10_000,
    resample_size=1.0,
    seed=1337,
)
result = comparator.compare()
print(f"P-Value: {result.pvalue}")
result.plot().show()

This project uses uv for dependency management. After installing uv, the dependencies can be installed using uv sync. This creates a virtual environment.

• The environment can be activated using source .venv/bin/activate (linux) or .venv/Scripts/activate
• Alternatively, commands can be run using, e.g., uv run app.py without activating the environment manually
• A requirements file can be generated using uv export > requirements.txt