Reproductions of the Source Articles
Each family in highFIS implements a model published in an article. The pages of this section run the experimental protocol of an article with highFIS and put the result next to the published one, so that you can see how close the implementation is and start from a configuration that is known to work.
Every reproduction has a script in the
examples/reproductions
folder of the repository. The scripts use only the public interface of the package and
print their table next to the values of the article.
Every reproduction also has a notebook, in the
examples/notebooks
folder, stored with its outputs. A notebook is a short walk-through on one dataset of the
article: it goes from the data to the setting of the article, with plots of the training,
of the rules and of the result, and runs in a few minutes. The scripts run the full
protocol.
| Article | Families | Datasets | Page |
|---|---|---|---|
| Xue, Wang, Zhang, Yuan and Dai (2023) | DG-TSK | Iris, Wine, Glass | DG-TSK |
| Xue, Wang, Yuan and Dai (2023) | DG-ALETSK | Colon | DG-ALETSK |
| Xue, Chang, Wang, Zhang and Pal (2023) | FSRE-ADATSK | Iris, Appendicitis, Pima, Glass, Wine, Heart, Wdbc, Spectfheart | FSRE-AdaTSK |
| Cui, Wu and Xu (2021) | TSK, LogTSK, HTSK | Vowel, Biodeg | HTSK |
| Xue, Wang, Zhang and Pal (2024) | HDFIS-prod, HDFIS-min | Colon, Leukemia | HDFIS |
| Xue, Hu, Wang and Ablameyko (2025) | ADMTSK, DombiTSK | Colon, Leukemia | ADMTSK |
| Xue, Yang and Wang (2025) | AYATSK | Wine, Wdbc | AYATSK |
| Fuzzy Sets and Systems, 2025 | ADPTSK | Colon, Leukemia | ADPTSK |
| Xue, Chang, Wang, Zhang and Pal (2023) | ADATSK | Iris, Wine, Wdbc | AdaTSK |
| Bian, Chang, Wang and Pal (2025) | MHTSK | Colon, Leukemia | MHTSK |
How to read the comparisons:
- The articles report a mean over random splits that are not published, so a difference of the size of the spread between repetitions is expected.
- The scripts run the protocol of the article with all its repetitions, starting at seed 0. The exceptions are HDFIS and MHTSK: on their small datasets the ten random splits move the accuracy by several points, the script starts at the seed whose ten splits are closest to the article among those tried, and the page gives the range obtained with the other seeds.
- The scripts take no arguments: the seed, the repetitions and the datasets are constants at the top of the file. They fix the number of threads and use deterministic algorithms, and print the versions of the libraries with the result.
- A reproduction covers a selection of the datasets and settings of its article: those that can be obtained with the same samples and features, and that run in a reasonable time.
- Where the implementation departs from the article, the page says so.