pytabkit

What it is and what it does

PyTabKit wraps modern tabular machine learning methods—neural networks like RealMLP and TabM, gradient boosted trees, and hybrid models—behind scikit-learn-compatible interfaces. It automatically handles GPU selection, categorical column detection, numerical preprocessing, and train-validation splitting for early stopping, with optional cross-validation ensembling and hyperparameter optimization.

The package is built around a NeurIPS 2024 paper benchmarking these methods on tabular classification and regression tasks. It includes both the trained model interfaces and the benchmarking code used to evaluate them. While the documentation recommends AutoGluon for best-possible results in production, PyTabKit is positioned as easier to use for development and supports vectorized cross-validation that can accelerate training of certain models.

Use it for:

• Quick prototyping of tabular classifiers and regressors with automatic preprocessing and GPU acceleration.
• Benchmarking multiple tabular ML methods on your own datasets using the built-in benchmarking utilities.
• Training RealMLP or TabM models with cross-validation ensembling and hyperparameter optimization for improved accuracy.
• Comparing gradient boosted tree implementations (XGBoost, LightGBM, CatBoost) with neural network baselines on structured data.
• Post-hoc calibration and refinement stopping for probabilistic predictions using temperature scaling.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

PyTabKit provides scikit-learn interfaces to modern tabular machine learning models—including RealMLP, gradient boosted trees (XGB, LGBM, CatBoost), and TabR—with built-in hyperparameter optimization, cross-validation ensembling, and benchmarking utilities.

Yes, with conditions. Install if you are developing tabular ML pipelines and want scikit-learn-compatible interfaces to modern methods with minimal setup friction. The low install overhead and automatic preprocessing are valuable for iteration. However, if your goal is production-grade best performance, the documentation explicitly recommends AutoGluon instead. Maintenance is aging but not stale; no known vulnerabilities.

Install

pytabkit on PyPI

pip install pytabkit

uv add pytabkit

poetry add pytabkit

Installing pytabkit

Before you install

Low install friction; pure Python wheel. Requires torch, pytorch-lightning, and scikit-learn as core runtime dependencies. Status is aging (220 days since last release), though the repository remains active and well-maintained with recent commits and no archived status.

License in practice

Apache-2.0 permissive license allows commercial and private use with minimal restrictions; suitable for most production and research contexts.

Quickstart

pip install pytabkit

from pytabkit import RealMLP_TD_Classifier

model = RealMLP_TD_Classifier()
model.fit(X_train, y_train)
model.predict(X_test)

Missing numerical values are not allowed and must be imputed beforehand. Requires torch to be installed separately if you need to control CPU/GPU version. TabR requires manual faiss installation (conda only).

Verify before relying

• Whether vectorized cross-validation speedup for RealMLP is significant enough to justify adoption over AutoGluon for production use cases.
• Current performance comparison with TabArena's newer models and preprocessing capabilities for real-world datasets.
• GPU memory requirements and scalability limits for large tabular datasets.

Package facts

Evidence: pytabkit-1.7.3-py3-none-any.whl

Keywords: RealMLP, deep learning, gradient boosting, scikit-learn, tabular data

Tags

More Artificial Intelligence packages