Today, many companies are immersed in the development of machine learning and artificial intelligence (AI). And it’s no secret that the development of machine learning operations have become widespread in recent years. Companies are fighting for the best specialists in data processing and analysis and machine learning engineers, pursuing one goal: to create great value using the power of AI.

Looking for solutions for your company?          Estimate project

In this guide we will discuss the MLOps consulting, what it is used for, the differences between MLOps and DevOps, as well as the various stages of the MLOps life cycle. You will find out the real-life examples of companies that have revealed their potential and value by implementing MLOps methods.

What is MLOps?

Let’s get started with the concept of MLOps. MLOps stands for Machine Learning Operations. MLOps is a new approach of collaboration between business representatives, machine learning specialists and IT engineers when it comes to developing artificial intelligence systems. In other words, it is a way of transforming machine learning operations and technologies into a useful tool for solving business problems.

“MLOps is an ML engineering culture and practice that aims at unifying ML system development (Dev) and ML system operation (Ops).” — Google

machine learning operations

The adoption of these methods improves quality, simplifies the management process, and automates the deployment of machine learning operations and deep learning models in large-scale production environments. The complete MLOps process includes three key steps: 

Dex-approximate application design: At this stage, you identify your potential user, develop a machine learning solution and analyze the further development of the project.

ML development: The second stage is characterized by testing the applicability of ML to the problem by implementing Proof-of-Concept for the machine learning model. The main goal is to provide a stable quality machine learning model that can be used in production.

ML Operations: At the third stage, it is important to deliver a previously developed machine learning model to production using proven DevOps methods.

The difference between MLOps and DEVops

Since MLOps were derived from DevOps principles, DevOps and MLOps have fundamental similarities. However, they still have differences in execution:

Continuous Integration (CI) is not only testing and validation of code and components, but also data, data schemas and models.

Continuous Delivery (CD) becomes even more difficult. It is not about a single service or package, but about whole complicated systems with models that are supported as services.

continuous delivery Source: medium.com

The models must be monitored, and the monitoring data must be used for recurrent training. Another issue is that models should be restored automatically without the need for external intervention.

Continuous Testing (CT) is an extra part that involves model testing and validation of models, which depends on the problem being solved, it is no longer just integration tests or unit tests.

How to implement MLOps?

If you are thinking about implementing MLOps, then there are 3 options how you can do it:

• MLOps level 0 (Manual process)
• MLOps level 1 (ML pipeline automation)
• MLOps level 2 (CI/CD pipeline automation)

MLOps level 0 (Manual process)

This level is typical for companies that are just getting started with machine learning. Each step in each pipeline is completed manually, including data preparation and validation, model training and testing. A completely manual machine learning workflow will be sufficient if your models are rarely modified or trained.

MLOps implementationSource: medium.com

MLOps level 1 (ML pipeline automation)

A key factor of MLOps level 1 implementation is to perform continuous learning (CT) of the model by automating the ML pipeline. This implementation option can be useful for companies that operate in a constantly changing environment and need to proactively reduce changes in customer behavior, price rates and other indicators.

machine learning operationsSource: towardsdatascience.com

MLOps level 2 (CI/CD pipeline automation)

With the help of the automated CI/CD system, data processing and analysis specialists can quickly explore new ideas related to function design, model architecture and hyperparameters. This level is appropriate for technology companies that have to retrain their models on a daily basis, update them in minutes and re-deploy them on thousands of servers simultaneously.

mlops implementationSource: towardsdatascience.com

The real-life use cases of MLOps in business

Ocado

About: Ocado is one of the world’s largest online supermarkets, whose systems process millions of user actions every minute on the company’s website and applications.

MLOps solutions: The company has implemented MLOps for fraud detection. Thanks to this technology, the firm can effectively ensure the legitimacy of the transaction when placing an order through the order management system.

Revolut

About: British financial technology company that offers banking services to its customers.

example of machine learning operationsSource: Building a state-of-the-art card fraud detection system in 9 months | by Dmitri Lihhatsov | Revolut Tech | Medium

MLOps solutions: Revolut has implemented MLOps technology for offline verification of millions of transactions and combating fraudulent card transactions. Using Google Cloud Stack driver and Kibana, Revolut monitors operational and functional performance indicators such as number of alerts and frauds, true positive rates (TPR), and false positive rates (FPR).

DoorDash

About: American company that operates an online food ordering and food delivery platform –

MLOps solutions: DoorDash uses MLops technology in several cases that are designed to optimize the work of users, sellers and consumers. The company has a variety of advantages as a result of implementing machine learning models:

• Forecasting and balancing of the offer (sellers) with the demand (of consumers) at any given time.
• Estimation of delivery time when a customer places an order
• Dynamic pricing
• Sellers’ recommendations to consumers
• Search ranking of the best sellers for DoorDash

machine learning Source: DoorDash’s ML Platform – The Beginning | DoorDash Engineering Blog

Conclusion: Investing in MLOps consulting

Wondering if your company needs MLOps?

Today, machine learning is one of the most significant breakthrough technologies that has huge opportunities for business transformation and digitization.

With MLOps consulting, your company will be able to create reproducible workflows and models, to easily deploy high precision models in any location and many more. Addepto is professional and experienced AI experts that can support your company in the field of consulting and implementation of MLOps solutions. Don’t hesitate to contact us!

 

 

 

Grow your businness with machine learning and big data solutions.

Our team of experts will turn your data into business insights.

growth illustration

Planning AI or BI project? Get an Estimate

Get a quick estimate of your AI or BI project within 1 business day. Delivered straight to your inbox.