Hands-On Machine Learning with Azure

Creating a predictive experiment

Before a trained model can be exported to the web service, the training experiment that created the model must be converted into a predictive experiment. A predictive experiment defines the scoring pipeline for creating predictions based on web service input. It does not contain any training modules, since the model is already trained. Instead, the model is just loaded from the list of trained models and imported as input to the Score Model module.

To create a predictive experiment, open a training experiment that has been previously run, or run the training experiment once to create a trained model. The experiment does not need to include a Score Model module—this will be added to the predictive experiment ...