Most autoML packages aim for exceptional performance but need to train for an exceptional amount of time. Fast-autoML aims for reasonable performance in a reasonable amount of time. In my work, I find this is a great tool for fast iteration in initial stages of data exploration.
Anecdotally, I’ve also found that fast-autoML performs as well as or better than autoML packages (auto-sklearn, TPOT, MLBox) for many smaller datasets that I work with.
Fast-autoML includes additional utilities, such as tools for comparing model performance by repeated cross-validation.