Counterfactual Explanations for Keras

This application is a simple example of using Conterfactual Explanations (Python) component to identify the Counterfactual Instances for a Binary classification model trained in Keras outside KNIME eg: Python IDE, Google Colab or Jupyter Notebook.

The Python object reader load the pickled Python object which is used for nomalisation of features, while the Network Reader node reads the keras model trained outside of KNIME. The component (in blue) can be used to select the instances to be used for Counterfactual Explanations.

External resources

• Counterfactual -Interpretable-ml-book
• KNIME Keras Integration
• How to set up Python-KNIME extension
• Python Library for Counterfactual Instances