Improve model explainability for created xgboost model

[GautamR-Samagra]

Automatically prepare datasets from user /item metadata and test interactions and relationships with synthetic data to provide better model explainabilty

[GautamR-Samagra]

Let's first look at how XGBoost models are usually represented and check if scala models for the same have some quick and easy representation techniques.

[GautamR-Samagra]

Show feature importance for all features extracted from the XGBoost model

We can create side by side graphs :

1. Items rank distribution for a new user (scatter plot) :

   • Items along with one selected item features
   • Ranks for the items on X axis

2. Feature distribution for selected items on the first graph :

   • show small histograms for all features on the other side