Code and Data for: Better by default: Strong pre-tuned MLPs and boosted trees on tabular data [NeurIPS, arXiv v2] (doi:10.18419/darus-4555)

View:

Part 1: Document Description
Part 2: Study Description
Part 5: Other Study-Related Materials
Entire Codebook

Code and Data for: Better by default: Strong pre-tuned MLPs and boosted trees on tabular data [NeurIPS, arXiv v2]

doi:10.18419/darus-4555

DaRUS

2024-11-05

1

Holzmüller, David; Grinsztajn, Léo; Steinwart, Ingo, 2024, "Code and Data for: Better by default: Strong pre-tuned MLPs and boosted trees on tabular data [NeurIPS, arXiv v2]", https://doi.org/10.18419/DARUS-4555, DaRUS, V1

Code and Data for: Better by default: Strong pre-tuned MLPs and boosted trees on tabular data [NeurIPS, arXiv v2]

doi:10.18419/darus-4555

Holzmüller, David (INRIA - Institut National de Recherche en Informatique et Automatique)

Grinsztajn, Léo (INRIA - Institut National de Recherche en Informatique et Automatique)

Steinwart, Ingo (Universität Stuttgart)

Strecker, Katharina

Dockès, Jérôme

EXC 2075 - 390740016

2023-AD011012804R1

2024-AD011012804R2

DaRUS

Holzmüller, David

Holzmüller, David

Grinsztajn, Léo

Steinwart, Ingo

Holzmüller, David

2024-10-29

https://doi.org/10.18419/DARUS-4555

Computer and Information Science, Tabular Data, Gradient Boosting, Benchmark, Artificial Neural Network

Artificial Intelligence and Machine Learning Methods

This dataset contains code and data for our paper "Better by default: Strong pre-tuned MLPs and boosted trees on tabular data", specifically, the NeurIPS version which is also the second version on arXiv. The main code is provided in pytabkit_code.zip and contains further documentation in README.md and the docs folder. The main code is also provided on <a href=https://github.com/dholzmueller/pytabkit>GitHub</a>. Here, we additionally provide the data that is generated by the code as well as the plots. See the documentation in docs/source/bench/download_results.md in the main code for instructions on how/when to download which data, or the documentation hosted <a href=https://pytabkit.readthedocs.io/en/latest/bench/download_results.html>here</a>. The code for the old version of the Grinsztajn et al. (2022) benchmark is provided in grinsztajn_benchmarking_code.zip and on <a href=https://github.com/LeoGrin/tabular-benchmark/tree/better_by_default>GitHub</a>. The code and data for the first arXiv version of the paper are archived <a href=https://doi.org/10.18419/darus-4255>here</a>.

David Holzmüller, Léo Grinsztajn, and Ingo Steinwart. Better by Default: Strong Pre-Tuned MLPs and Boosted Trees on Tabular Data, Neural Information Processing Systems, 2024.

2407.04491

David Holzmüller, Léo Grinsztajn, and Ingo Steinwart. Better by Default: Strong Pre-Tuned MLPs and Boosted Trees on Tabular Data, Neural Information Processing Systems, 2024.

Label:

grinsztajn_benchmarking_code.zip

Text:

Code for running the old version of the Grinsztajn. et al (2022) benchmark with our models.

application/zip

Label:

grinsztajn_results.csv.gz

Text:

Results on the old version of the Grinsztajn et al. (2022) benchmark.

application/gzip

Label:

pytabkit_code.zip

Text:

Contains the code for running the benchmarks, as well as an implementation of the models evaluated on these benchmarks. Also contains instructions in README.md as well as the documentation.

application/zip

Label:

main_no_results.tar.gz

Text:

Benchmark results data for all benchmarks. Contains result summaries (enough for plotting) but not the detailed results. After unpacking, rename the tasks_only_infos folder to tasks if you don't already have a tasks folder.

text/plain

Label:

results_small.tar.gz

Text:

Results folder (within the data folder) without extra results (predictions on the datasets, optimal hyperparameters). After unpacking, the folder should be renamed to "results".

text/plain

Label:

tasks.tar.gz

Text:

Folder with imported datasets, with restricted access for copyright reasons.

text/plain

Label:

CatBoost-HPO_steps.tar.gz

Text:

text/plain

Label:

cv_refit.tar.gz

Text:

Results for inner cross-validation / refitting of RealMLP-TD and LGBM-TD.

text/plain

Label:

FTT-HPO_steps.tar.gz

Text:

text/plain

Label:

LGBM-HPO_steps.tar.gz

Text:

text/plain

Label:

MLP-PLR-HPO_steps.tar.gz

Text:

text/plain

Label:

MLP-RTDL-HPO_steps.tar.gz

Text:

text/plain

Label:

RealMLP-HPO_steps.tar.gz

Text:

text/plain

Label:

ResNet-RTDL-HPO_steps.tar.gz

Text:

text/plain

Label:

results_main.tar.gz

Text:

Detailed results, including predictions on datasets and best parameters, for all main methods. This excludes the data for individual hyperparameter optimization (HPO) steps, which is provided separately.

text/plain

Label:

RF-HPO_steps.tar.gz

Text:

text/plain

Label:

TabR-HPO_steps.tar.gz

Text:

text/plain

Label:

XGB-HPO_steps.tar.gz

Text:

text/plain