remove redundant features + train with xgboost

[salelkafrawy]

Here is a new feature selection way:

• Inspecting what features could be redundant and manually removing them
• Tune an XGBoost model on the remaining features
• Train the tuned model
• Add the new model to generate_figures.py
• This model beats the NN one on AUC
• Here is the comet url

[TimkLee]

The notebook produces some very cool images! The scripts seems fine and ran for me without any problem. It is interesting that all these models have a very similar performance.