wwydmanski / hypertab

Code for "HyperTab: Hypernetwork Approach for Deep Learning on Small Tabular Datasets"

HyperTab

HyperTab is a hypernetwork-based classifier for small tabular datasets.

It's especially efficient when the number of samples is smaller than 500. The smaller the dataset, the larger is the advantage of HyperTab over other algorithms.

Installation

pip install hypertab

Usage

from hypertab import HyperTabClassifier
DEVICE = "cuda:0" if torch.cuda.is_available() else "cpu"

clf = HyperTabClassifier(0.5, device=DEVICE, test_nodes=100, epochs=10, hidden_dims=5)
clf.fit(X, y)
clf.predict(X)

Performance

Dataset	XGBoost	DN	RF	HyperTab	Node
Wisconsin Breast Cancer	93.85	95.58	95.96	97.58	96.19
Connectionist Bench	83.52	79.02	83.50	87.09	85.61
Dermatology	96.05	97.80	97.21	97.82	97.99
Glass	94.74	46.96	97.02	98.36	44.90
Promoter	81.88	78.91	85.94	89.06	83.75
Ionosphere	90.67	93.43	92.43	94.52	91.03
Libras	74.38	81.54	77.42	85.22	82.72
Lymphography	85.94	85.74	87.19	83.90	83.93
Parkinsons	86.35	74.96	86.84	95.27	80.20
Zoo	92.86	72.62	92.62	95.27	89.05
Hill-Valley without noise	65.53	56.39	57.33	70.59	52.71
Hill-Valley with noise	58.45	56.06	55.66	67.56	51.09
OpenML 1086	60.61	33.33	51.24	76.60	68.39
Heart	79.17	82.62	81.10	83.33	82.38
Mean rank	3.50	3.78	3.07	1.35	3.29

hypertab's Issues

Regression for HyperTab?

Hi Witold,

I've been using your HyperTab classification algorithm and am impressed with its performance. I'm curious if there are plans to add a regression feature to handle continuous data. This enhancement would greatly expand its usability for various datasets.

In the paper, you mentioned that HyperTab can be applied to regression tasks, too. Is this feature on your roadmap?

Thanks for your great work on this project.

Best,
Mustafa Akben