1. Fact: The reality of most ML modeling on data is a lot of models don't need to be giant ass interaction machines with 10^80 features. Many non linear models

1. Fact: The reality of most ML modeling on data is a lot of models don't need to be giant ass interaction machines with 10^80 features. Many non linear models with a Naive Bayes or other simple non parametric models and adding a few Boolean Random Indicator Variables to your regression will go a long way to make your models non linear and get you to 80% F1 score. These are all practical problems and none of them require just a pure DS but also ML skills 2. Fact: You need to deal with bigger problems in ML like data set class imbalance, calibrating responses to the right scalar range(figuring out what that range even is in terms of domain). This is not taught in schools, just like writing Software is not taught in schools. One needs to be in the field to learn these skills and an ML Engineer can pick these up as much as a DS.