In this example, the values for height would most likely have more than two possible values, and hence we might choose Multiclass Classification or even Regression. Again, it would be best, since the model builder makes quick work of this, to try both and then evaluate the results.
Moving on, if you indicated a manual approach (or selected the Manual mode) to build the model, you'll have to click on Add Estimators (in the upper right of the page) to choose one or more specific estimators. Choosing the right model estimator is often even harder than choosing the technique. Different estimators are a better match for different types of data and for solving different problems.
With the model builder, based upon the classification technique chosen, you'll have a number of estimators to choose from (as shown in the following screenshot). For example, we chose Multiclass Classification, so you could pick one of the following estimators:
- Decision Tree Classifier
- Random Forecast Classifier
- Naive Bayes
Here, we will choose the Random Forest Classifier estimator and click on Add:
After clicking Add, and then Next, the model will be trained on the data, technique, and estimator selected. The results will then be posted to the Select model page (as shown in the following screenshot), where you can click on Save to save the model results for evaluation:
A Watson Machine Learning model (created with the Model Builder) becomes an asset and is listed as such (shown above) on the IBM Watson Studio project page for later reference, refinement and reuse.
It is worth mentioning that an approach to learning how to select a classification technique (algorithm) and estimator is through experimentation with the model builder. In other words, using the broader terms of classification and feature selection, the model builder now makes it effective and efficient enough to test hypotheses with a variety of approaches, easily assess the results, and then deploy the best fit (more on evaluating model performance and accuracy in a later section of this chapter and throughout this book) as a continuous learning model with new and unseen data.
We will see more on this process with some experimentation later in this chapter.