Make cache for trained models

[Sum02dean]

This prevents the need to retrain each time a new feature is implemented.

Perform for all three models: xgboost bambi neural-net