Using evaluations to increase the effectiveness of your Personalized Service

After you have deployed the service and are collecting data about user interactions and using the rank and reward system, you may want to assess ways to enhance your model, Actions, Context, or Features. Rather than making changes to the current configurations, Microsoft gives you the option to assess these options by using offline evaluations, assuming that enough data has been captured to do so. Offline evaluations allow you to test your features and how effective they’ve been over the span of your learning loop. With this option, you can specify a date range for your testing through the current time without affecting the existing model or the performance of the model for users interfacing with the application. As discussed previously in the chapter, if you make changes to certain configurations, it will redeploy your model, and you will need to build it from scratch and lose all the history amassed...