ML model management

A machine learning lifecycle refers to the multiple steps needed in order to build and maintain a ML model.

A sub-area of machine learning life cycle is Model Management that contains experiment tracking, versioning, deployment and the hardware scaling.

Mlflow

Mlflow cares about model management just that, no production monitoring or data preparation. Even the free version only cares about till model versioning.

Using Folder system as a very basic way of managing model versions ?

1. Error Prone You name folders yourself, and maybe you end up with renaming a model version or delete it
2. No versioning things become messy when the number of versions increase, as the number of folders increase with the version number
3. No model lineage you just save the model without knowing about its HyperParameters, Metrics, or training and validation data sets.