Model

For detailed information on the specifications of each model, please see Generalized Linear Model. Tutorials for each algorithm, and model specification through R, are available in Tutorial Documentation.

In general, a model of the user’s choosing can be specified either by finding the list of algorithms at the top of the Inspect page when data are parsed or by selecting the appropriate model from the drop down menu Model.

Each model requires that the user provide a .hex key associated with a data set. Users can often begin typing the name of the original data source, and select the appropriate .hex key from the auto fill menu that appears. Users can also find .hex keys for data sets by selecting View All from the Data drop down menu, or for all H2O actions by selecting Jobs from the Admin drop down menu.

If a large data set is used in the training and testing of a model, H2O’s capabilities can be bounded by the amount of memory available on the machine. To utilize H2O’s full capability, the amount of memory available should be about 4 times the file size of the data set, but not more than the machine’s total available memory. For instructions on how to change the amount of memory allocated to H2O see the Quick Start Documentation. Advanced users should run H2O on a cloud computing resource or server.

Grid Search Models

GLM, and GBM both offer Grid Search Models. In order to access this option in GLM uses should select GLM Grid Search from the Model drop down menu.

Each grid search modeling option allows users to generate multiple models simultaneously, and compare model criteria directly, rather than separately building models one at a time. Users can specify multiple model configurations by entering different values of tuning parameters separated by coma. For example, to specify three different values of lambda, a regularization parameter in GLM Grid search users might enter: .001, .05, .1.

When multiple values are specified for many tuning parameters grid search returns one model for each unique combination. For example, in GBM, if users specify Ntrees as 50, 100, 200, and also specify learning rates of 0.01, and 0.05, six models will be returned.

Grid search results return a table showing the combination of tuning parameters used for each model and basic model evaluation information, as well as a link to each model. Users can access the details of each model by clicking on the model links in the table.