Before presenting MLFlow capabilities, here it is the documentation page for further usage:

MLflow Documentation

Important concepts

• ML Experiment: the process of building a ML model
• Experiment run: each trial in an ML experiment
• Run Artifact: any file associated with an ML experiment

What is experiment tracking?

It is the process of keeping tracking of relevant information regarding an experiment, like:

• source code
• logs
• data
• environment settings
• model
• metrics

MLFlow

It is an open-source package that allows machine learning models management. It is divided into 4 parts: Tracking, Models, Model Registry and Projects. The focus of this bootcamp is on Tracking and Model Registry.

Inside some important features

An important step before saving any run in a specific experiment is to create an experiment! Here is how we can do this:

import mlflow
mlflow.set_experiment('experiment_name')

Done! From now on, every run will be saved inside this experiment. But as we are trying many combinations of model parameters and these trials might affect model performance metric, it will be very useful to log every detail of our trials for comparisons. How can we do it?