Model-driven Stochastic Trace Clustering

Description

This is the github repository containing the code, experiments and experimental results used in the paper:

Model-driven Stochastic Trace Clustering Jari Peeperkorn, Johannes De Smedt & Jochen De Weerdt

Using the code

The folder entroclus contains the code for the clustering algorithm.

alternatives contains all the alternative clustering algorithms used in the experiments (with the exception of ActiTraC).

tutorials to be added

Structure

├── entroclus/                      # The main files needed for our model-driven clustering apporach
│   ├── entropic_clustering.py          # The script used for running our method
│   ├── entropic_clustering_variants.py # Every alternative of entroclus
│   ├── entropic_clustering_utils.py    # Utilities such as initialization
│   ├── entropic_relevance.py           # Custom python script to run our version of ER
│   └── utils.py                        # Containing extra utilities such as DFG discovery
├── alternatives/                   # Alternative clustering algorithms (except ActiTraC)
│   ├── frequency_based.py              # Frequency-based clustering
│   ├── random_clustering.py            # Random clustering
│   ├── trace2vec_based.py              # Trace2Vec-based clustering
│   └── trace2vec_fixed.py              # Trace2Vec with fixed parameters
├── evaluation/                     # Evaluation metrics and utilities
│   ├── graph_simplicity_metrics.py     # Graph simplicity metrics
│   ├── metrics.py                      # General metrics
│   ├── stochastic_PN.py                # Stochastic Petri Net evaluation
│   └── tade_conformance.py             # TADE conformance metrics
├── experiment/                     # All scripts and files needed for the experiment + output
│   ├── datasets/                        # Datasets used in the experiments
│   ├── experimental_results_elbow/      # Results for elbow experiments
│   ├── experimental_results_one_cluster_size/ 
|   |   ├── clusters/                    #Contains the DFGs for each of the clusters we mined
|   |   ├── results/                     #Results for fixed cluster size experiments                  
│   ├── evaluate_clusters_one_cluster_size.py  # Evaluation script
│   ├── get_average_ranks.py             # Compute average ranks
│   ├── get_average_ranks_statistical_test.py # Statistical tests on ranks
│   ├── get_clusters_one_cluster_size.py # Get clusters for fixed size
│   ├── get_elbow_results.py             # Get elbow method results
│   ├── get_log_stats.py                 # Log statistics
│   ├── plot_elbow_experiment.py         # Plotting elbow experiment results
│   ├── reproduce_experiments.py         # Script to rerun all experiments
│   ├── run_elbow_experiment.py          # Run elbow experiment
├── requirements.txt                # Python dependencies
├── README.md                       # This file

Reproducing the experiment

1. Install the required dependencies:

pip install -r requirements.txt

2. To reproduce the main experiments, run the following script:

python experiment/reproduce_experiments.py

This will execute all experiments and generate the results in the experiment/experimental_results_elbow/ and experiment/experimental_results_one_cluster_size/ directories.

Citing

If you use this code or results in your research, please cite the following paper:

Model-driven Stochastic Trace Clustering Jari Peeperkorn, Johannes De Smedt & Jochen De Weerdt

Contact

For questions or issues, please contact Jari Peeperkorn (jari.peeperkorn@kuleuven.be).