Overview

DeepTab is a Python library that brings modern deep learning architectures to tabular data. It wraps PyTorch and Lightning behind a scikit-learn-compatible interface, so you can use state-of-the-art models without changing how you already work with data.

Why DeepTab

Tabular data is the most common format in applied machine learning, yet most deep learning tooling is designed for images or text. DeepTab fills that gap by:

Providing a consistent fit / predict / evaluate API across all models.
Handling categorical encoding, numerical preprocessing, and batching automatically.
Supporting regression, classification, and distributional regression from the same model class.
Integrating with scikit-learn pipelines and hyperparameter search tools.

Available models

All models support regression, classification, and distributional regression out of the box. Import them as <ModelName>Regressor, <ModelName>Classifier, or <ModelName>LSS.

Stable

Model	Architecture	Reference
`Mambular`	Sequential Mamba (SSM) blocks for tabular data	Thielmann et al. (2024)
`MambaTab`	Mamba block on a joint input representation	Ahamed et al. (2024)
`MambAttention`	Mamba + Transformer hybrid	Thielmann et al. (2025)
`FTTransformer`	Feature tokeniser + Transformer encoder	Gorishniy et al. (2021)
`TabTransformer`	Transformer with categorical embeddings	Huang et al. (2020)
`SAINT`	Row attention + contrastive pre-training	Somepalli et al. (2021)
`TabM`	Batch ensembling for MLP	Gorishniy et al. (2024)
`TabR`	Retrieval-augmented tabular model	—
`ResNet`	ResNet adapted for tabular data	—
`MLP`	Multi-layer perceptron baseline	—
`NODE`	Neural oblivious decision ensembles	Popov et al. (2019)
`NDTF`	Neural decision tree forest	Kontschieder et al. (2015)
`TabulaRNN`	Recurrent neural network for tabular data	Thielmann et al. (2025)
`ENODE`	Extended NODE variant	—
`AutoInt`	Automatic feature interaction via attention	—

Experimental

Experimental models are imported from deeptab.models.experimental. Their API may change without a deprecation cycle. See Using experimental models for a worked example.

Model	Architecture	Reference
`ModernNCA`	Modern neural classification architecture	—
`Trompt`	Tabular-specific prompting model	—
`Tangos`	Tabular model with graph-based structure	—

Next steps

Installation — install DeepTab and verify the setup.
Key Concepts — understand the API patterns before writing code.
Examples — runnable end-to-end workflows.
API Reference — full parameter documentation.