MLOps design

1. MLOps design

Good work on finishing the first chapter, where we focused on a high-level overview of the different phases. We will now take a more detailed look into the design phase.

2. Machine learning design

We will now look into the start of the design phase, in which we investigate the impact, business requirements, and the key metrics that we need to track.

3. Added value

It's very useful to estimate the value a machine learning model is expected to bring into our business, since machine learning development is experimental and uncertain. This aids in resource allocation, project prioritization, and setting realistic expectations, ensuring effective and responsible ML development and deployment.

4. Business requirements

Apart from the expected value of the machine learning model, we must also consider the business requirements. Let's say we are building a machine learning model that predicts the number of sales of a specific product, such that we can purchase the right amount to put in our shop. The machine learning model will output a predicted amount of sales. We must consider the frequency of the predictions and how fast we need them. We must also evaluate the model's accuracy and whether its results are explainable to non-experts. Transparency is a crucial factor in machine learning, since it enables us to find why the model makes its predictions, why it is wrong, and how we can improve the model. These requirements all have an impact on what algorithm we will use.

5. Business requirements

Depending on the problem we are trying to solve, there could also be compliance and regulatory requirements to the usage of machine learning. For instance, in finance, when the law requires an explanation from the system.

6. Business requirements

We should also take the available budget and team size into account.

7. Key metrics

To see if the machine learning lifecycle progresses as expected, it is often wise to track the performance of the model. However, as we’ve seen earlier, the roles involved in MLOps processes are multidisciplinary and thus also have their own way of tracking performance. The data scientist looks at the accuracy of a model, how many times the algorithm is correct.

8. Key metrics

The subject matter expert is interested in the model's impact on the business, for instance, how their work improves due to the use of machine learning. They are primarily interested in domain-specific metrics.

9. Key metrics

The business stakeholder is more interested in the monetary value of the model, in how many cases we actually generate revenue. This is often expressed in money or time. To get the most out of machine learning, we must align the different metrics to make sure everyone is on the same page.

10. Let's practice!

Now that we've looked into the design phase of the machine learning lifecycle, let's go into some exercises.

Create Your Free Account

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