ML Flow is an open source tool by Databricks to track the complete Machine Learning project lifecycle.
Consider that you have a project at hand, and you make 20 different models with different parameters and metrics and other details.
How do you keep a track of all that? How do you make sure that switching to a different model is easy?
Now one way could be to maintain an Excel sheet and enter all the model details, hyperparameters, metrics, etc. But this is clearly not feasible for a full scale project.
Here’s where a tool like ML Flow comes in handy, and it can be integrated with all the major ML libraries easily. In fact, Pycaret has got ML Flow UI as an integrated feature.
ML Flow currently has 4 main components to it:
ML Flow Tracking: is what I described above, an interactive UI for logging and tracking model details.
ML Flow Projects: is a code packaging format to enable sharing of code and reproducability using Conda and Docker.
ML Flow Models: is a model packaging service which lets you easily deploy and score the same model on different platforms like Sagemaker and Azure.
ML Flow Model Registry: is a centralized API to collaboratively manage the full lifecycle of MLflow Models.