Learning about MLflow
MLflow is an open source project from Databricks that helps to manage ML life cycles. Databricks provides a completely managed and hosted version of MLflow. The features of MLflow include the following:
- Tracking: Data scientists train several models at a time, and it can be hard to track them all. MLflow makes it easy to track the trained models using different algorithms, by providing a logging mechanism. We can even compare different results and parameters.
- Artifacts: The models can be packaged in reusable forms and shared with other data scientists and ML engineers.
- Registering: MLflow allows us to register different ML models and even different versions of these models. This makes it easy to manage an ML model's entire life cycle and allows for transitioning.
- Deployment: The models registered using MLflow can be hosted as REpresentational State Transfer (REST) endpoints. This allows for easy model querying and deployment.
MLflow...