Code and experiments for the paper "Process discovery on deviant traces and other stranger things"

This repository include the data and the support code to run the experiments, as well as the code implementing the algorithm described in the paper (in the dist directory).

Running the experiments

The experiments should be run using Jupyter, either using its web based interface or nbconvert. The required software can be installed using Anaconda Python distribution or manually installing the clingo ASP solver (version 5.4.0 must used!). We strongly suggest the former because it's easier to ensure a reproducible environment via virtual environments. The required conda packages are listed in the binder/environment.yml file (on Windows the environment.win.yml file must be used instead).

Below you'll find specific instructions for different options to set-up the Jupyter working environment, once the environment is running you can just open the run_*_experiments.ipynb and run every cell. Alternatively, on Linux or OSX, you can open a command line terminal with the right conda environment and run the command:

jupyter nbconvert --ExecutePreprocessor.timeout=-1 --to html --execute run_*_experiments.ipynb

that executes all the notebooks and saves their output in separate HTML document (on Windows the command line might need to be adapted).

The raw data generated by the experiments will be available in the results directory.

Run locally

Assuming that Anaconda or Miniconda are already installed you can just create the virtual environment and start Jupyter (or use the nbconvert workflow without starting the server):

conda env create -f=binder/environment.yml
conda activate negdis
jupyter lab

Using Binder

The repository is compatible with Binder and repo2docker. To use repo2docker on the local machine you can use the command jupyter-repo2docker with the path of the cloned repository or its URL. E.g.:

jupyter-repo2docker .

which starts a container with Jupyter that can be accessed via web browser.

Using both these methods the filesystem of the container will be separated from the one on the host, so the results must be exported using the Jupyter interface. Jupyter enables the use of a command line terminal that can be used for nbconvert.

Using Docker

You can also use Docker with a Jupyter image, it's similar to the Binder option, but you could mount the local filesystem on the container, so you don't need to copy the results afterwards. First you need to start a container with Jupyter that can access the local directory:

docker run --rm -p 8888:8888 -v "$PWD":/home/jovyan/work jupyter/minimal-notebook

The local directory will be available in the work directory within the Jupyter interface. Then you need to install the required packages opening a terminal (within the Jupyter interface) and run the command:

conda env update -n=base  -f=work/binder/environment.yml

For details see Jupyter Docker Stacks - Running a Container. The -v option mounts the current directory on the virtual machine, so the results of the experiments will be stored on the local filesystem.

Available experiments

Due to missing distribution rights for the CERV dataset, the files with the event logs might not be included in this repository. Please get in touch with the authors for more details.

ASPrin optimisation experiments

• run_asprin_cerv_experiments: CERV dataset
• run_asprin_dreyers_experiments: Dreyers dataset

Experiments presented in Process discovery on deviant traces and other stranger things

• run_cerv_experiments: CERV dataset
• run_reallife_experiments
• run_synt_experiments