What is MLOps?

1. What is MLOps?

Hello, nice to meet you! I am Arne and I will be your instructor for this course! MLOps is a promising emerging field that can leverage the business value of machine learning.

2. Course Outline

We will start in Chapter 1 by explaining what MLOps is, why businesses will benefit from adopting it, and what is required before implementing MLOps. Chapter 2 will discuss the MLOps life cycle. In Chapter 3, we will learn how to do MLOps in practice, and in the final chapter 4, we will discuss best practices and pitfalls and conclude with a real case study.

3. Prerequisites

Before taking MLOps for Business, we advise completing the prerequisite course. Otherwise, many of the concepts discussed here might not be familiar to you, and you will not learn as much as expected.

4. What is MLOps?

MLOps stands for machine learning operations. Operations means putting a machine learning model into production. It is derived from DevOps, an established and successful set of practices that combines traditional software development (the Dev part of DevOps) with IT Operations (the Ops part). The aim is to reduce friction between those who develop and those who operate IT products and speed up delivery time, increase reliability, and improve collaboration. We will learn more about what DevOps is and which practices make it successful, but for now: MLOps is the application of DevOps principles for data science and machine learning development and operations.

5. What is MLOps (continued)?

MLOps is a set of concepts, tools, and best practices. It is not yet well-defined since it is still an emerging field with the general aim to standardize, streamline, and automate both the development and operation of machine learning models. What does this mean? Successful MLOps will help you to scale from operating a handful of machine learning applications to hundreds or potentially even thousands of them.

6. Supply chain forecasts

A topic I worked on is demand forecasting. The goal is here to get a good idea of how many different products we will sell in the future. Imagine you start with a few manual monthly forecasts for your company's key products. Each of these products might initially use different models. Starting with the standardizing, we run each possible model on each product and compare the predictive power to choose the best. This helps us to scale the forecasting. Scaling can mean that we forecast not only our key products but all of them. Scaling can also mean that we deliver weekly, perhaps even daily, forecasts of future demand. Automating means that we will not trigger the forecasts manually anymore, and not humans, but automated tests will ensure that the product forecasts work as intended.

7. How is MLOps used

Other applications include, among others, automated translation tools such as Google Translator. With translation tools, the challenge is not only to develop a decent model to translate a text automatically, or even a spoken word, into another language. It is also to make these models usable on a cellphone with limited computational capacity in a subway with background noise, perhaps even without an Internet connection. Other examples include automotive cars or recommendation engines: Netflix, for example, estimates that 80% of its video selections come from automated machine-learning recommendations. They also include the automated supervision of production processes, something I implemented in my previous job, or search engines. All these examples have in common that the machine learning models operate automatically, as in the case of automated cars, perhaps in charge of potentially life-threatening split-second decisions. This is MLOps.

8. Benefits of MLOps

A well-designed MLOps application will provide scalability, exactly moving from one model to hundreds of them, automation, which is a prerequisite for scaling, and an important way to avoid human-introduced errors, stability or reliability, for example, minimizing downtimes as well as maintainability or adaptability, which ensures that we can do maintenance work and adapt to changing business requirements.

9. Let's practice!

In the next video, we will look at the business case and the requirements for MLOps, but before that, let's practice!

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.