Use it when
- You want to organize projects and runs and track your experiments (manual and automatic logging), artifacts, and data.
- You want to keep track of your models with a model registry and serve them using integrations.
- You want a platform that is non-opinionated and gives you flexibility.
Watch out
- MLflow can track data but provides limited capability in terms of data versioning. You may have to integrate other tools.
- MLflow's built-in model serving is quite limited. You will likely need to integrate with a third-party tool for a robust solution.
Example stacks
Airflow + MLflow stack
Installation
pip install mlflow