MLflow

You can check out all the detail of MLflow from this website https://mlflow.org/. Anyhow, we know you are lazy to go through the document yourselve. Thus, we will try to explain and give a brief guide to kick-start your MLflow journey.

• What is MLflow?
• Why?
• How to use MLflow?

What is MLflow?

Quote from link

MLflow is an open source platform to manage the ML lifecycle, including experimentation, reproducibility, deployment, and a central model registry. MLflow currently offers four components:

1. MLflow Tracking - Record and query experiments: code, data, config, and results.
2. MLflow Projects - Package data science code in a format to reproduce runs on any platform.
3. MLflow Models - Deploy machine learning models in diverse serving environments.
4. Model Registry - Store, annotate, discover, and manage models in a central repository.

In more simple words. First, MLflow provides a tensorboard likes website for you to track your ML/DL training. Second, MLflow helps you to log your experiments. Third, MLflow is a centralized service to deploy ML/DL models.

Why?

Well. You can imagine running/developing/experimenting ML/DL models. In your project, you might need to try on 3 - 4 models and each model requires changing 10 parameters. How would you log your result? How about that one time you randomly try a new parameter and it just works?

This is just one reason why we need to use MLflow.

How to use MLflow?

For AIT BrainLab, the MLflow is deployed at LA. You can access the MLflow via http://la.cs.ait.ac.th. (please note that the link starts with http, not https).

An example MLflow can be found here