The life-cycle of Machine Learning models and Software Artifacts such as pipelines, etc needs to be managed to implement MLOps practices. In this chapter of the MLOps tutorial, you will learn about the Management of Machine Learning models and Software Artifacts.
The management of Machine Learning models and Software Artifacts revolves around the Versioning and Documentation of Machine Learning models and Software Artifacts.
The Problem
As the number of Machine Learning and Data Science departments in a company scales, there are more models and pipelines being used for various tasks- improving existing models, creating new models for new tasks, etc. The conventional approach to managing Machine Learning models and pipelines quickly starts to become a laborious and grueling task, a nightmare. The amount of manual effort required for managing the machine learning models is both unscalable and unproductive.
The Solution
To tackle this problem, proper version control and documentation are required for each and every Machine Learning model and Software artifact. This includes keeping track of the changes in the model over time(such as those caused by retraining the model on the latest data), the data used for training and re-training the model. Alongside, keeping track and storing of various metrics of Machine Learning models such as accuracy, precision, etc along with business specific KPI(Key Performance Indicators) is also of paramount importance when managing Machine Learning models.
This management of models is not only applicable to models that are in production but also to models that are currently in the various stages of the pipeline. Management and tracking of models in various stages can help in accelerating the development of models by helping various teams and departments of the organization to collaborate.
For the optimal advantage of the organization, it is a good practice to manage all the Machine Learning models and Software Artifacts centrally. The benefits of which are-
- Helps in Auditing.
- Makes it possible to use/ reuse the models by other teams and departments in the organization.
- Helps with Governance and Compliance.
- Makes Reporting and Monitoring easier.
- Makes it easy to collaborate between various departments and teams of an organization.
Further, proper versioning and documentation reduce the risk factors associated with Machine Learning models, more on that in a later chapter.
The documentation and versioning of Software Artifacts are the same as any other piece of software, and won’t be discussed in this MLOps tutorial.