Machine Learning

Training Analytics

H2O

To create a new analytic using H20, right click on the datalist to be used for training and select New > Analytic.

In the new dialog, select H20 for the analytic type. Next, choose the algorithm to be used for training. For more information regarding individual algorithms, see the H20 documentation.

Next, choose the training view to be used when fitting the model.

Finally, choose the dependent (target field) and independent (select fields).

Modelshop and H20 will do their best to choose datatypes for you, however in the event it is appropriate, you may modify the type if any selected variable. This is particularly useful if your model target is a number, however you would prefer the model produce a discrete classification rather than regression. In this instance, you would change the target variable from an Interger, to an Enumeration.

Using an Analytic

You can easily predict with a trained algorithm by creating a new calculation and using the following syntax:

model.analytics.<Name of Analytic>.scoreMe()

The scoreMe() method will find the input values based on header names, so they should mirror the training data set.

Example: