Potential issue with Tree Surrogate rules as text

[dipanjanS]

Hi, as discussed when I was building my model interpretation tutorial on the census or adult dataset, the surrogate tree model throws an error when trying to get the rules as text.

Code and data I'm using is in my tutorial: http://nbviewer.jupyter.org/github/dipanjanS/data_science_for_all/blob/master/tds_model_interpretation_xai/Human-interpretable%20Machine%20Learning%20-%20DS.ipynb

Code causing an error (not in the tutorial but the surrogate model code is present)

Note: The model is in the variable surrogate_explainer in the code and works quite well otherwise for visualization and also predictions!

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-231-0d5c9d6ab4fe> in <module>()
----> 1 surrogate_explainer.decisions_as_txt()

C:\Anaconda3\lib\site-packages\skater-1.1.1b5-py3.6.egg\skater\core\global_interpretation\tree_surrogate.py in decisions_as_txt(self, scope, X)
    426         """ Retrieve the decision policies as text
    427         """
--> 428         tree_to_text(self.__model, self.feature_names, self.__model_type, scope, X)

C:\Anaconda3\lib\site-packages\skater-1.1.1b5-py3.6.egg\skater\core\visualizer\tree_visualizer.py in tree_to_text(tree, feature_names, estimator_type, scope, X)
    178     right_nodes = tree.tree_.children_right
    179     criterias = tree.tree_.threshold
--> 180     feature_names = [feature_names[i] for i in tree.tree_.feature]
    181     values = tree.tree_.value
    182 

C:\Anaconda3\lib\site-packages\skater-1.1.1b5-py3.6.egg\skater\core\visualizer\tree_visualizer.py in <listcomp>(.0)
    178     right_nodes = tree.tree_.children_right
    179     criterias = tree.tree_.threshold
--> 180     feature_names = [feature_names[i] for i in tree.tree_.feature]
    181     values = tree.tree_.value
    182 

TypeError: 'NoneType' object is not subscriptable

[SNEHARAMTEKE]

You can try adding the argument feature_names = list_of_your_feature_names in InMemoryModel. Doing this worked for me!