notesum.ai

Published at November 26

On the Efficiency of NLP-Inspired Methods for Tabular Deep Learning

cs.LG

Released Date: November 26, 2024

Authors: Anton Frederik Thielmann¹, Soheila Samiee²

Aff.: ¹BASF SE, Germany; ²BASF Canada Inc., Canada

Arxiv: http://arxiv.org/abs/2411.17207v1

Refer to caption

	Regression Tasks							Classification Tasks					Avg. Rank
Models	DI $\downarrow$	AB $\downarrow$	CA $\downarrow$	WI $\downarrow$	PA $\downarrow$	HS $\downarrow$	CP $\downarrow$	BA $\uparrow$	AD $\uparrow$	CH $\uparrow$	FI $\uparrow$	MA $\uparrow$
FT-Transformer*	0.018	0.458	0.169	0.615	0.024	0.111	0.024	0.926	0.926	0.863	0.792	0.916	2.25
	$\pm$ 0.001	$\pm$ 0.055	$\pm$ 0.006	$\pm$ 0.012	$\pm$ 0.005	$\pm$ 0.014	$\pm$ 0.001	$\pm$ 0.003	$\pm$ 0.002	$\pm$ 0.007	$\pm$ 0.011	$\pm$ 0.003
MLP*	0.066	0.462	0.198	0.654	0.764	0.147	0.031	0.895	0.914	0.840	0.793	0.886	5.58
	$\pm$ 0.003	$\pm$ 0.051	$\pm$ 0.011	$\pm$ 0.013	$\pm$ 0.023	$\pm$ 0.017	$\pm$ 0.001	$\pm$ 0.004	$\pm$ 0.002	$\pm$ 0.005	$\pm$ 0.011	$\pm$ 0.003
ResNet*	0.039	0.455	0.178	0.639	0.606	0.141	0.030	0.896	0.917	0.841	0.793	0.889	4.25
	$\pm$ 0.018	$\pm$ 0.045	$\pm$ 0.006	$\pm$ 0.013	$\pm$ 0.031	$\pm$ 0.017	$\pm$ 0.002	$\pm$ 0.006	$\pm$ 0.002	$\pm$ 0.006	$\pm$ 0.013	$\pm$ 0.003
Mambular*	0.018	0.452	0.167	0.628	0.035	0.132	0.027	0.927	0.928	0.856	0.795	0.917	2.00
	$\pm$ 0.000	$\pm$ 0.043	$\pm$ 0.011	$\pm$ 0.010	$\pm$ 0.005	$\pm$ 0.020	$\pm$ 0.002	$\pm$ 0.006	$\pm$ 0.002	$\pm$ 0.004	$\pm$ 0.011	$\pm$ 0.003
MambAttention	0.018	0.484	0.189	0.638	0.030	0.142	0.026	0.919	0.921	0.857	0.781	0.911	3.67
	$\pm$ 0.000	$\pm$ 0.052	$\pm$ 0.006	$\pm$ 0.003	$\pm$ 0.006	$\pm$ 0.024	$\pm$ 0.002	$\pm$ 0.004	$\pm$ 0.002	$\pm$ 0.004	$\pm$ 0.009	$\pm$ 0.001
TabulaRNN	0.018	0.459	0.178	0.659	0.073	0.114	0.027	0.930	0.925	0.855	0.796	0.922	2.75
	$\pm$ 0.000	$\pm$ 0.047	$\pm$ 0.013	$\pm$ 0.013	$\pm$ 0.012	$\pm$ 0.014	$\pm$ 0.001	$\pm$ 0.004	$\pm$ 0.002	$\pm$ 0.006	$\pm$ 0.011	$\pm$ 0.002
* adopted from Thielmann et al., (2024)