Building a Simple Deployment Workflow
So far, in all the case studies we have explored, we have always performed some kind of preprocessing of the input data, such as encoding categorical features, encoding text, or normalizing data, to name just some of the adopted preprocessing steps. During deployment, the new incoming data must be prepared with the exact same preprocessing as the training data in order to be consistent with the task and with the input that the network expects.
In this section, we use the sentiment analysis case study shown in Chapter 7, Implementing NLP Applications, as an example, and we build two deployment workflows for it. The goal of both workflows is to read new movie reviews from a database, predict the sentiment, and write the prediction into the database.
In the first example, the preprocessing steps are implemented manually into the deployment workflow. In the second example, the Integrated Deployment feature is used.
