Xplique is a Python toolkit for neural network explainability, offering a comprehensive suite of tools for understanding complex models. It targets researchers and practitioners in AI, providing methods to analyze model behavior, identify influential features, and extract human-understandable concepts.

How It Works

Xplique is structured into four core modules: Attributions Methods, Feature Visualization, Concepts, and Metrics. It implements state-of-the-art techniques for generating feature attributions (e.g., Grad-CAM, Integrated Gradients), visualizing internal representations, extracting human-interpretable concepts (e.g., CAV, CRAFT), and evaluating explanation quality with various metrics. The library supports both TensorFlow and PyTorch models, with a dedicated wrapper for PyTorch integration.

Quick Start & Requirements

• Install via pip: pip install xplique
• Requires Python 3.7+ and TensorFlow. PyTorch support is available via a wrapper.
• Official documentation: https://deel-ai.github.io/xplique/
• Tutorials: https://colab.research.google.com/ (links provided in README)

Highlighted Details

• Supports a wide range of tasks including Classification, Regression, Object Detection, and Semantic Segmentation.
• Handles various data types: Images, Time Series, and Tabular data.
• Includes an "Example-based" module for explanations using relevant data samples.
• Offers a PyTorch wrapper for seamless integration of PyTorch models.

Maintenance & Community

The project is actively developed by the DEEL project team, with contributions welcomed. Further information and community links are available in the README.

Licensing & Compatibility

Released under the MIT license, allowing for commercial use and integration with closed-source projects.

Limitations & Caveats

Some methods may not function as expected with Keras 3.X (TensorFlow 2.16+); TensorFlow 2.15 or earlier is recommended for optimal compatibility. The "Example-based" module is in its early stages.