MPPN

This repository contains the source code of the paper Multivariate Business Process Representation Learning for Predictive Process Analytics, which is currently under review for the BPM 2021.

Installation

The recommended way to install the scripts is:

cd mppn
conda create --name mppn python=3.7.
conda activate mppn
python setup.py install

This way all requirements are also installed including pytorch, cuda and fastai.

To run the jupyter notebooks you also have to create a jupyter kernel:

python -m ipykernel install --user --name mppn

Content

• 00_preprocessing.ipynb contains all the pre-processing logic for the process prediction tasks, including normalization and categorization, subsequence generation and data loader creation.
• 01_pipeline.ipynb includes the pipeline and the runner method to run all prediction models and tasks.
• 02_baselines.ipynb includes the implementation of 4 current approaches of process prediction.
• 03_mppn.ipynb presents the implementation of the MPPN that is described in the paper.
• 04_prediction_evaluation.ipynb runs the evaluation and logs the results.
• The four jupyter notebooks build up and are also exported to python files that are located under ./mppn/.
• The experiment can also be run from command line with python mppn/prediction_evaluation.py
• The event logs are stored under ./event_logs/
• The results are stored under ./results/

The repository is build with the nbdev package -> https://nbdev.fast.ai/.