MLOps Bootcamp

We define what MLOps as a methodology encompasses and why it is strategically important for forward-thinking companies to embrace it. You will get an overview of which organizational, technological and human aspects MLOps touches upon. We identify the challenges in these areas and use the principles of MLOps to show possible solutions to these challenges.

What aspects do you need to consider when starting an ML project? What can you do to ensure that established best practices are taken into account in your ML projects, even through the project structure? You will learn how to create and maintain this framework during the project. We look at, among other things:

• Standardized project structures
• Development workflows
• Reproducible deployment
• Code quality

In this module, we take a look at which automation tools are suitable for effectively automating repetitive tasks in ML projects. You will also learn what additional requirements ML projects place on already established CI/CD workflows from the DevOps environment and how to deal with them in practice. In the process, we clarify the following topics:

• How do you focus on the essentials of your project through automation?
• What repetitive tasks can you automate and how?
• Specifically, how do you turn complex manual tasks into simple workflows?
• How do you minimize the error rate through automation?

Clean data management is essential for ML projects. First of all, you will learn how to correctly manage and transparently version the corresponding data for your ML project. We then consider how you can ensure that this data always corresponds to the assumptions made in the development phase, even in ongoing operations. Here, we go through the essential stages of data management:

• Ingestion
• Validation
• Transformation
• Versioning

How can transposed ML models be versioned in the project? We analyze this and work out steps to ensure that the entire team is aware of which model is being put into production and how it is performing. At the same time, you will learn how to become audit-ready through this approach:

• Model authoring & model training
• Experiment management & experiment tracking workflow
• Model performance
• Model evaluation

There are several deployment models to get ML up and running. You will learn about the most important patterns and the differences between them, and how to orchestrate multiple models and the business logic you need. We also look at how ML models can be tested and optimized to keep latency and throughput at constant levels.

• Differences between offline/batch, asynchronous & synchronous patterns
• Business logic integration
• Multi model deployment

You will learn how to monitor models during operation with little effort, how to notice model, data and concept drift in time and which reactions are required in this case. We consider, among other things:

• Logging features & predictions
• The setup of a monitoring dashboard
• Dat, concept & model drift detection & handling

How can high test coverage be achieved quickly and easily? You will learn how to create unit tests for ML projects and what additional requirements arise from data and models. Topics include:

• Code coverage
• CVE checks
• Efficient collaboration through testing
• Unit tests with data and ML models