This repo demonstrates the applicability of the probabilistic learner DR-BART in Business Process Simulation (BPS) models. DR-BART was proposed in [Orlandi et al. (2021)[https://doi.org/10.48550/arXiv.2112.12259].
The DR-BART implementation is written in R and C, while we use Python to implement our BPS model. Therefore, this repo provides a wrapper to use trained DR-BART models from the implem 6B63 entation of Orlandi et al. in Python. Our wrapper code can be found in this file.
To test DR-BART for BPS, run the following steps:
run:
git clone [https://github.com/user/repo.git](https://github.com/ltsstar/TaskExecutionTimeMining)
cd TaskExecutionTimeMining
sh real_data_loader.sh
pip install -r requirements.txtWe use pipenv with pyenv. Make sure you have pipenv and pyenv installed and run:
pipenv install
pipenv shellTo preprocess the downloaded event logs, navigate to the src\notebooks directory, select the desired data set folder, e.g. BPIC_2017 and run the load Jupyter notebook.
E.g., for the BPIC 2017 data set run this notebook.
If you want to (re)train DR-BART models, navigate into the models directory and run the desired .sh files for training.
Run the evaluation Jupyter notebooks, e.g. this notebook.
Several demo Jupyter notebooks exist, e.g. this notebook.
This image shows that our DR-BART models correctly estimate with a low probability that a process will finish on weekends.