Tabnetviz

Tabnetviz: Open-Source Library for Interpreting Trained TabNet Models

An open-source library for explaining trained TabNet models, generating interactive reports to illustrate prediction explanations and feature importance scores.

What is Tabnetviz?

Tabnetviz is an open-source Python library that is designed to help interpret and explain TabNet models. TabNet is a neural network architecture for tabular data that can handle both classification and regression tasks. Tabnetviz generates interactive reports and visualizations that allow you to understand:

• Overall model performance
• Predictions for individual samples
• Global feature importance scores
• Feature influence on a per-sample basis

The interactive reports in Tabnetviz allow you drill down to see exactly how different features impact the predictions according to the model. This level of interpretability makes it easier to debug models, detect potential biases, and trust predictions for real-world usage. Tabnetviz supports TabNet models trained with PyTorch or TensorFlow backends.

Key capabilities include:

• Prediction explanation for regression and classification
• Partial dependence plots
• Decision tree approximations
• Feature influence maps
• Built-in reports for analyzing overall model and out-of-sample model performance

Tabnetviz is easy to use with just a few lines of code once a TabNet model has already been trained. It is a useful toolkit for any team leveraging TabNet for tabular data tasks where model interpretability is important.

Tabnetviz Features

Official Links