Introduction to Containers as Reproducible Models

Need for containers

When deploying data science models, it’s important to be able to reproduce the same environment used both for training and serving. In Chapter : Models as Web Endpoints, we used the same machine for both environments, and in Chapter 3 : Models as Serverless Functions we used a requirements.txt file to ensure that the serverless ecosystem used for serving the model matched our development environment. Container systems such as Docker provide a tool for building reproducible environments, and they are much lighter weight than alternative approaches such as virtual machines.