A Machine Learning Life Cycle, (abbreviated ML Life Cycle) is a conceptual framework of the cyclical process that ML projects go through. However, there is no agreement on what its various phases are. Nevertheless, the following elements are popular. In consecutive order:
- Understand the problem
- Collect the data
- Wrangle the data
- Develop the model
- Deploy the model
These are the visions of various data/AI companies:
- In this AWS blog post, the author drills down from a very high-level overview of the ML life cucle to a very detailed view. (⚠️recommended)
- In this blog post, an engineer at Google focusses a lot on the underlying technology of the ML life cycle. It only uses three phases: planning, engineering and modeling.
- DataOps company Neptune has a very extensive write-up on six different phases in the ML life cycle.
- This conceptual framework by DataRobot adds an extra phases: interpreting & communicating the model results.