H2O AutoML
The most efficient method of model building and tuning utilizes H2O AutoML. AutoML builds models from multiple algorithms while implementing appropriate grid search and model optimization based on the model type. The user can specify constraints such as compute time limits or limits on the number of models created.
Some features of AutoML include the following:
- AutoML trains a random grid of GLMs, GBMs, and DNNs using a carefully chosen hyperparameter space.
- Individual models are tuned using a validation set or with cross-validation.
- Two stacked ensemble models are trained by default: All Models and a lightweight Best of Family ensemble.
- AutoML returns a sorted leaderboard of all models.
- Any model can be easily promoted to production.
Stacked ensembles are highly predictive models that usually appear at the top of leaderboards. Similar to the other ensemble approaches that we introduced earlier (such as bagging and boosting), we stack works...
