How to use LIME on h2o GBM model to do feature selection of columns in Python

Question

Trying to get the top 3 columns from my h2o GBM model using lime which has the highest significance in model prediction.

Answer 1

There is no out-of-the-box solution yet in open source h2o-3, but there are many examples of how to do this available. Here are the repos/notebooks:

• https://github.com/jphall663/interpretable_machine_learning_with_python / https://github.com/jphall663/interpretable_machine_learning_with_python/blob/master/lime.ipynb

• https://github.com/h2oai/mli-resources / https://github.com/h2oai/mli-resources/blob/master/notebooks/lime.ipynb

• https://content.oreilly.com/oriole/Interpretable-machine-learning-with-Python-XGBoost-and-H2O / https://content.oreilly.com/oriole/Interpretable-machine-learning-with-Python-XGBoost-and-H2O/blob/master/lime.ipynb

There is also some chance Marco Tulio's original LIME package will work: https://github.com/marcotcr/lime, be sure to look into this example: https://marcotcr.github.io/lime/tutorials/Tutorial_H2O_continuous_and_cat.html