DeepTLF: A Framework for Enhanced Deep Learning on Tabular Data

Overview

DeepTLF significantly outperforms traditional Deep Neural Networks (DNNs) in handling tabular data. Using our novel TreeDrivenEncoder, we transform complex, heterogeneous data into a format highly compatible with DNNs. This enables a 19.6% average performance increase compared to conventional DNNs.

Installation

You can install DeepTLF directly from PyPI:

pip install deeptlf

Quick Start

Seamlessly integrate DeepTLF into your workflow through its scikit-learn-compatible API:

from deeptlf import DeepTFL

# Initialize and train model
dtlf_model = DeepTFL(n_est=23, max_depth=3, drop=0.23, n_layers=4, task='class')
dtlf_model.fit(X_train, y_train)

# Make predictions
dtlf_y_hat = dtlf_model.predict(X_test)

Features

• Transforms heterogeneous data into DNN-friendly format
• Supports multimodal learning
• Adheres to the scikit-learn API for effortless integration
• Features advanced options like custom layers, dropout rates, and more

Citation

To cite DeepTLF in your work:

@article{borisov2022deeptlf,
  title={DeepTLF: robust deep neural networks for heterogeneous tabular data},
  author={Borisov, Vadim and Broelemann, Klaus and Kasneci, Enkelejda and Kasneci, Gjergji},
  journal={International Journal of Data Science and Analytics},
  pages={1--16},
  year={2022},
  publisher={Springer}
}