This chapter will provide you with the skills and knowledge needed to move your machine learning models from the research and development phase into a production environment. You will learn about the process of moving from a research prototype to a reliable, scalable, and maintainable system.

Adopting an MLOps mindset

What is a key aspect of MLOps?

What is technical debt?

Writing maintainable ML code

Version control

Code organization

Writing effective ML documentation

Why document labeling methods?

MLOps best practices

Moving from Research to Production

In this chapter, you’ll learn about the importance of reproducibility in machine learning, and how to ensure that your models remain reproducible and reliable over time. You’ll explore various techniques and best practices that you can use to ensure the reproducibility of your models.

Designing reproducible experiments

ML reproducibility

Identifying Key Aspects for Reproducibility and Reliability

Sorting Reliability Elements

Sorting Aspects of Reproducibility and Documentation Practices

Feature engineering

Classifying Feature Engineering Goals & Techniques

Identifying Data Manipulation and Transformation Techniques

Feature selection

Data and model versioning

Model store

Versioning training data

Ensuring reproducibility

 Ensuring Reproducibility

In Chapter 3, you’ll examine the various challenges associated with deploying machine learning models into production environments. You’ll learn about the various approaches to deploying ML models in production and strategies for monitoring and maintaining ML models in production.

Packaging ML models

Goal of serialization and containerization

Model packaging order

Scalability

Scaling

Model complexity

Categorizing scaling techniques

Automation

Benefits of continuous delivery

CI/CD steps

Putting ML models into production

ML in Production Environments

In the final chapter, you’ll learn about the various ways to test machine learning pipelines and ensure they perform as expected. You’ll discover the importance of testing ML pipelines and learn techniques for testing and validating ML pipelines.

Model reliability

Options

Testing Pipeline Readiness

Testing for Pipeline Staleness

Categorizing Testing Strategies for ML Pipeline Readiness and Staleness

Addressing Model Staleness in Machine Learning Pipelines

Testing data

Picking the right test

Consistency check

Testing models

Predicting customer lifetime value

Individual Fairness

Group Fairness

Classifying fairness

Testing Model-producing Code

Testing Data

Testing Models

Testing machine learning pipelines

Congratulations!

Testing ML Pipelines

Much of today’s machine learning-related content focuses on model training and parameter tuning, but 90% of experimental models never make it to production, mainly because they were not built to last. In this course, you will see how shifting your mindset from a machine learning engineering mindset to an MLOps (Machine Learning Operations) mindset will allow you to train, document, maintain, and scale your models to their fullest potential.

Much of today’s machine learning-related content focuses on model training and parameter tuning, but 90% of experimental models never make it to production, mainly because they were not built to last. In this course, you will see how shifting your mindset from a machine learning engineering mindset to an MLOps (Machine Learning Operations) mindset will allow you to train, document, maintain, and scale your models to their fullest potential.
Experiment and Document with Ease
Experimenting with ML models is often enjoyable but can be time-consuming. Here, you will learn how to design reproducible experiments to expedite this process while writing documentation for yourself and your teammates, making future work on the pipeline a breeze.
Build MLOps Models For Production
You will learn best practices for packaging and serializing both models and environments for production to ensure that models will last as long as possible.
Scale Up and Automate your ML Pipelines
By considering model and data complexity and continuous automation, you can ensure that your models will be scaled for production use and can be monitored and deployed in the blink of an eye.

Once you complete this course, you will be able to design and develop machine learning models that are ready for production and continuously improve them over time.

MLOps Concepts

Supervised Learning with scikit-learn

Shift to an MLOps mindset, enabling you to train, document, maintain, and scale your machine learning models to their fullest potential.

Developing Machine Learning Models for Production with an MLOps Mindset

Developing Machine Learning Models for Production

Principes de base des MLOps

Same credit score and income, but different interest rates

Two candidates have extremely similar qualifications and experience but one is rejected in the very first round and the other gets the job.

A healthcare model is less likely to recommend useful treatments for people working in factories because they might incur too many costs.

Online ads that show high-paying job opportunities more often to men than women

Less accurate for people with darker skin tones

Police using facial recognition technology that is less accurate for certain types of individuals, resulting in wrongful arrests

Classifying fairness

Developing Machine Learning Models for Production

Exercice interactif pratique