This repository contains to code to evaluate different classification algorithms on eight benchmark problems from engineering design. In particular, it evaluates the novel pretrained classification model: TabPFN.
The datasets used to evaluate the various classification algorithms are available
here. Download the whole data
folder and
put it at the root of this repository.
The datasets are located in data/processed
, while the performance of the considered
classifiers are in data/results
. The files are in Arrow format (parquet) and
are best read with pandas
.
To recreate the plots of our paper, you can run the plot.ipynb
notebook.
If you use the datasets or the code for research purposes, you can cite our paper:
Cyril Picard and Faez Ahmed, Fast and Accurate Zero-Training Classification for Tabular Engineering Data, arXiv:2401.06948.