ML asset deployment
As a data engineer, you will be tasked with building out an ML model deployment process. This way, science can be made part of a solution and then operationalized. Even if that deployment is to a user acceptance testing (UAT) environment, from the perspective of the data scientist, it is operationalized. So, the first configuration option you have to support is the deployment to a target environment. You must be able to un-deploy (rollback) any ML model as a package with equal ease. You need to then follow a zero-footprint rule – that is, leave no trace after removal.
There will be many scripted and parameterized configurations (or steps) to be taken when deploying ML models. You want to build a deployment framework or use a proven third-party tool. Being able to call the deployment process from a notebook is also really useful. Some tools and MLOps frameworks used for model deployment are as follows:
- Amazon SageMaker {https://packt-debp.link...
