Now is a good time to have a look at how you can test the model prediction right here in IBM Watson Studio. To do this, you can click on Test:

The default test format generated shows you an input form that you can use to enter data values. Later, you'll see that if you have an external process generating test data, you can use the input format icons to use JSON data file format and paste in your data test values:

For now (staying on the Test tab), leave the default format (input form) and enter some values for the important columns (the input data form is populated with a sample record from the dataset). To test the model, change the values and click on Predict:

1. For year, enter 2016
2. For position, enter QB
3. For weight, enter 225
4. Click on Predict

Once your model test is complete, IBM Watson Studio displays a graphical score (with percentages) of the column's ability to predict the result. In the following example, fields 5 and 6 are position and weight, respectively, and we can see by the model's performance scores, their ability to predict the result:

So, given the preceding output, we can perhaps conclude that in our NFL player statistical file, a player's position is a pretty good indicator as to what the player's height is.

If you are so inclined, you can click the output formatting icon (shown circled in the following screenshot) and convert the performance information to raw output (View raw output):

When you deploy your model in this way (using Watson Studio deploy), the deployment is a one-time event. In other words, you enter your data, train the model on the data, then see the performance results and draw your conclusions.

This works for exploration and investigation purposes, but realistically, you'd want to retain and continually train the model with new data as it becomes available. To do this, you can use the IBM Watson Machine Learning continuous learning system, which provides automated monitoring of model performance (discussed briefly in the next section of this chapter), retraining, and redeployment to ensure prediction quality from your model.