Xplique 0.1.0

Explanations toolbox for Tensorflow 2

Xplique is a Python module dedicated to explainability. It provides several submodules to learn more about your tensorflow models (≥2.1). The three main submodules are Attributions Methods, Explainability Metrics and Feature Visualization tools.

The Attributions Method submodule implements various methods, with explanations, examples and links to official papers.

Soon, the Explainability Metrics submodule will implement the current metrics related to explainability. These evaluations used in conjunction with the attribution methods allow to measure the quality of the explanations.

Soon, the Feature Visualization submodule will allow to represent neurons, channels or layers by maximizing an input.

The package is released under MIT license.

Contents

• Install
  
• Get started
  
• Core features
  
  • Attributions Methods
    
  • Concept based Methods
    
  • Metrics
    
  • Feature Visualization
    
• Notebooks
  

Installing

The library has been tested on Linux, MacOSX and Windows and relies on the following Python modules:

• Tensorflow (>=2.1)
• Numpy (>=1.18)

You can install Xplique using pip with:

pip install xplique

Getting Started

let's start with a simple example, by computing Grad-CAM for several images (or a complete dataset) on a trained model.

from xplique.attributions import GradCAM

# load images, labels and model
# ...

method = GradCAM(model)
explanations = method.explain(images, labels)

Notebooks

• Using the attributions methods

Core features

Attributions Methods

• Deconvolution 📚^Api 📄^arxiv
• Grad-CAM 📚^Api 📄^arxiv
• Grad-CAM++ 📚^Api 📄^arxiv
• Gradient Input 📚^Api 📄^arxiv
• Guided Backprop 📚^Api 📄^arxiv
• Integrated Gradients 📚^Api 📄^arxiv
• Occlusion 📚^Api 📄^arxiv
• Rise 📚^Api 📄^arxiv
• Saliency 📚^Api 📄^arxiv
• SmoothGrad 📚^Api 📄^arxiv
• SquareGrad 📚^Api 📄^arxiv
• VarGrad 📚^Api 📄^arxiv
• Ablation-CAM
• Xray

Concept-based Methods

• Concept Activation Vector [^12]
• Testing with Concept Activation Vector [^12]
• Robust TCAV
• Automatic Concept Extraction

Metrics

• Aocp
• Fidelity correlation
• Irof
• Pixel Flipping
• Stability

Feature Visualization

• Vanilla