Building a completed solution

For the first proof of concept, we decided to use the Ocean Smart seafood catering business unit to display additional recommendations to accompany selected items and present a user with a list of platters to select from. The goal is for the first item in the list to be the selection that the user is most likely to purchase. To accomplish this, we create a function app with two functions – GetRankedActions and RewardAction. The GetRankedActions action presents the list of possible Actions for the user, ordered by the probability that the user will purchase that item. RewardAction tells the Personalizer service whether that item was purchased or not. To train the model, we have a test script that selects a user at random, presents that user with choices from the Personalizer, and lets the Personalizer know whether the first choice presented was one the user would purchase. Here is a basic diagram of the flow:

Figure 12...