Data can be downloaded from zenodo: https://zenodo.org/record/5532578 (Note: we added a new version since 2021/10/04). Unzip the files and you should get a data and results folder in the root directory of the repo.
Several scientific python packages are required to run the code. The easiest way to do this is to use the python 3.X Anaconda distribution (https://www.continuum.io/downloads).
- jupyter
- scipy
- pandas
- numpy
- statsmodels
- scitkit-learn
- matplotlib
- Cython
- git+https://github.com/jfrelinger/cython-munkres-wrapper
- fisher
The Cython code for the evaluation metrics should also be compiled. For this, change directory to /lib and run python setup.py build_ext --inplace
The Cython munkres library should be installed using e.g. pip install git+https://github.com/jfrelinger/cython-munkres-wrapper
After installation, run jupyter notebook in the notebooks/ folder. This folder contains several Jupter notebooks, each pertaining to a different part of the evaluation study. The notebooks starting with 0- are optional if you downloaded the results folder.
Code was tested on python 3.7 to 3.9