This repository contains the code for SDFastEst and its evaluation. Hence, it contains the sampling framework, code for processing the original datasets, code to collect algorithms' data, and code for application examples.
applications
: Contains code for three different use-case of SDFastEst (runtime meta-learning, warm starting, spectrum plots).data
: Contains result data of SDFastEst; Plots from the paper (and any other relevant plot); Code to collect the search space size for different Pysubgroup algorithms, the resulting data of the collection (i.e., the ground truth used for evaluation), all SD task settings (combinations of depth, dataset, target value); Moreover, it contains the code used to process datasets, an overview of all used datasets.general_utils
: Contains general utility code like file management. Also contains the code used to plot all plots in the paper and produce the tables' data.SDFastEst
: Contains code to run the sampling framework SDFastEst, the framework itself, and auxiliary code like algorithm or sampling profiles.sdfastest_comparer.py
: Main file used to produce SDFastEst's estimations and results. Setup to produce the results used in the paper plus some additional results with other sampling profiles.
- Download and preprocess the datasets
- Set up the config.py accordingly to your own environment
- Set up the Python environment with all requirements as listed in the requirements.txt
Having done the above, you can test some applications, run the fastest_comparer.py or directly use the SDFastEst framework.
If you use our code or data in scientific publications, we would appreciate citations.
Estimating the Pruned Search Space Size of Subgroup Discovery, Purucker et al., International Conference on Data Mining, 2022
Link to publication.
@inproceedings{purucker2022sdfastest,
author = {Lennart Purucker and
Felix I. Stamm and
Florian Lemmerich and
Joeran Beel},
title = {Estimating the Pruned Search Space Size of Subgroup Discovery},
booktitle = {{IEEE} International Conference on Data Mining, {ICDM} 2022, Orlando,
FL, USA, November 28 - Dec. 1, 2022},
year = {2022},
}