Putting it all together – algorithms, feature engineering, grid search, and AutoML

The H2O AutoML implementation is simple yet powerful, so why would we ever need grid search? In fact, for a lot of real-world enterprise use cases, any of the top candidates in an AutoML leaderboard would be great models to put into production. This is especially true of the stacked ensemble models produced by AutoML.

However, our coverage of grid search was not just to satisfy intellectual curiosity. A more involved process, which we will outline next, uses AutoML followed by a customized grid search to discover and fine-tune model performance.

An enhanced AutoML procedure

Here are the steps:

1. Start by running AutoML on your data to create a baseline leaderboard. You can investigate leading models, gain an understanding of the runtimes required to fit algorithms to your data, and more, which may inform future AutoML parameter choices and expectations.
2. The second stage is feature...