LimeWrapper provides a simple wrapper around the lime package. Lime provides local explanations of the predictions of h2o or sklearn models. While the lime documentation already provides tutorials to do this, the goal of this wrapper is to abstract away the feature engineering processes required for lime. This, however, requires model re-training if you have categorical features (no re-training required if all features are continuous). If that's resource or time intensive, then my recommendation would be directly following the tutorials by the lime authors available in their github repository.
Please refer to the Jupyter notebook in the example folder to see a classification and regression model explanation example.
Clone the repository and cd into it.
To install package dependencies run:
pip install --trusted-host pypi.python.org -r requirements.txtTo install the limewrapper package run:
python setup.py installFor small bug-fixes and new features, create a separate branch, make necessary commits and create a pull request.
- Support for sklearn models.
- Function to get top n explainable features for entire test set.