Overview

In this workshop, participants will learn about MLflow, an open source platform for the complete machine learning life cycle. The focus of the workshop will be on how to use MLflow to organize, track and reproduce research experiments, ensuring the reproducibility of results and enabling collaboration among research teams.

Motivations

As a researcher practitioner, one faces numerous challenges when organizing and tracking experiments. Experimenting on different machines, with varying configurations and dependencies can lead to discrepancies in results and difficulties in reproducing results. Keeping track of multiple experiments and determining the best approach to a problem can also be a cumbersome task.

MLflow presents a solution to these challenges by providing a centralized platform for tracking experiments, organizing code and reproducing results. This makes research more efficient and trustworthy.

  • The process of organizing and tracking research experiments becomes streamlined and simplified with MLflow.
  • A centralized platform for logging information related to models, such as parameters, code versions and results, is provided by MLflow.
  • MLflow makes it easy to track down previous parameters and figures used in experiments, as all the relevant information is stored in one place.
  • MLflow abstracts away the maintenance of a database, allowing researcher practitioners to focus on logging important information without being burdened by technical details.
  • Similarly it abstracts away the maintenance of artifacts (data, figures), saving them under a common structure associated with each experimental run.

MLflow’s ability to simplify and streamline the process of organizing and tracking research experiments makes it an indispensable tool for researcher practitioners.

Learning Outcomes

At the end of this workshop, participants will be able to:

  • Set up and use MLflow to track experiments
  • Use MLflow to organize code and reproducing results
  • Compare and reproduce results from multiple experiments
  • Collaborate with research teams by sharing experiments and results

Lesson Outline

  1. Introduction to MLflow and its components
  2. Setting up MLflow and tracking experiments
  3. Logging code and results in MLflow
  4. Reproducing experiments and results
  5. Comparing experiments and results in MLflow
  6. Collaborating with research teams using MLflow

Workshop Format

  1. Presetantation: The workshop will consist of a 45-minute presentation that covers the basics of MLflow and its benefits for organizing research experiments.
  2. Practice: An optional multi-hour hour workshop where participants will walk through the process of deploying MLflow alongside an artifact store and database by using docker-compose. This hands-on experience will give participants a practical understanding of how to use MLflow in their own research projects.

See the workshop structure page for more information about general workshop format.