MLflow Training Course

Introduction

• Adapting software development best practices to machine learning for government.
• Evaluating MLflow versus Kubeflow — where does MLflow excel?

Overview of the Machine Learning Cycle

• Data preparation, model training, model deployment, and model serving, among other steps.

Overview of MLflow Features and Architecture

• MLflow Tracking, MLflow Projects, and MLflow Models.
• Utilizing the MLflow command-line interface (CLI).
• Navigating the MLflow user interface.

Setting up MLflow for Government

• Installation in a public cloud environment.
• Deployment on an on-premise server.

Preparing the Development Environment

• Working with Jupyter notebooks, Python integrated development environments (IDEs), and standalone scripts.

Preparing a Project

• Establishing connections to data sources.
• Creating a prediction model.
• Training the model.

Using MLflow Tracking

• Logging code versions, data sets, and configurations.
• Recording output files and performance metrics.
• Querying and comparing experimental results.

Running MLflow Projects

• Overview of YAML syntax for configuration.
• The role of Git repositories in version control.
• Packaging code for reusability and scalability.
• Sharing code and collaborating with team members to enhance productivity.

Saving and Serving Models with MLflow Models

• Selecting an environment for deployment (cloud, standalone application, etc.).
• Deploying the machine learning model in a secure and efficient manner.
• Serving the model to ensure real-time or batch processing capabilities.

Using the MLflow Model Registry

• Setting up a central repository for models.
• Storing, annotating, and discovering models for reuse.
• Collaboratively managing models to ensure consistency and accountability.

Integrating MLflow with Other Systems

• Working with MLflow plugins to extend functionality.
• Integrating with third-party storage systems, authentication providers, and REST APIs.
• Optional integration with Apache Spark for big data processing.

Troubleshooting

Summary and Conclusion