There are two workflows: One for data which is not large. Full (or 80% of data) can be used for training and rest for testing, feature generation and subsequent ML model building. IInd workflow is for large data where a fraction of it is used for 'autofeat' modeling and then this model is used to create features in the full training dataset. Two components have been used: auto feature generator I and auto feature generator II.
This is a companion discussion topic for the original entry at https://kni.me/w/rdAE3FyNIWrbZgYa