I’ve used TPOT several times and while it does have functionality for feature preprocessing, such as Polynomial Features, or feature selection such as Principal Components Analysis, it does not have automated feature engineering. TPOT requires only a single table of data and cannot make the deep features that Featuretools is able to by combining feature primitives.

As far as I’m aware, Featuretools is the only open-source Python library for the kind of automated feature engineering done in these projects. If you want to use TPOT with multiple tables, you’ll first have to make a feature matrix with Featuretools.

TPOT can work well for model optimization, but you’re going to need a lot of computing resources and a very long time to allow the genetic algorithms time to evolve an optimal model. I think that it’s a great library and would encourage people to check it out or other automated modeling libraries such as auto-sklearn.