What is MLOps and why is it important?

by Bart Marseille, on Feb 17, 2021 1:35:20 PM

Blog Bart Marseille, regarding MLOps.

Bart Marseille

Bart, what is your role at Crystalloids?

For over 10 years, I have been a Data Scientist and Machine Learning Engineer/Researcher at Crystalloids. My role is to extract meaning and insights from data by developing and applying machine learning systems and quickly prototyping new ideas/technologies related to ML.

What is MLOps?

ML can be a game changer for a business, but without some form of systemisation, it can devolve into a science experiment. The real challenge isn't building an ML model, the challenge is building an integrated ML system and to continuously operate it in production.

MLOps is a practice for collaboration and communication between data scientists and operations professionals to help manage the production ML lifecycle. It is an ML engineering culture and practice that aims at unifying ML system development (Dev) and ML system operation (Ops).

The diagram below shows that only a small fraction of a real-world ML system is composed of the ML code. The required surrounding elements are vast and complex, requiring a team of Data Scientists, developers and operations professionals working together effectively and systematically.

What is MLOps and why is it important?

Figure 1. Elements for ML systems. Adapted from Hidden Technical Debt in Machine Learning Systems.

Why is MLOps being adopted more and more?

MLOps brings business interest back to the forefront of the ML operations. Data scientists work with the business interest and goals in mind, with clear direction and measurable benchmarks. 

MLOps follows a similar pattern and principles to DevOps and DataOps. The practices that drive integration between the development cycle and the overall operations process can also transform how the business handles data. Just like DevOps shortens production life cycles by creating better products with each iteration, MLOps drives insights you can trust and put into play more quickly in a controlled manner.

What can it be used for and give us some examples

ML is very powerful: it’s a process that generalises from data examples. It provides the ability to understand or value something/someone quite new to you. But ML itself is a means to an end. It’s what you do with that understanding/value, the business value, that will make the difference. If it fits in your businesses strategy, you should use MLOps to achieve this. Some examples we operate at Crystalloids:

A. Predict customer/visitor conversion likelihood and use it for e.g. bid optimisation.

Measures like Return on Ad Spend (ROAS) and Profit on Ad Spend (POAS) allow you to optimise to spend your money on visitors that are likely to convert and not to target ones less likely to convert.

How do you identify which customers are very likely to convert? By using predictive modelling I try to exploit the behavioural patterns that precede conversion. The past behaviour of visitors, combined with their conversion information can be applied to future visitors. For example, it could be that customers who recently visited the website with many page recurring views  are much more likely to buy something in the coming 15 days. To improve that likelihood even further, targeting them with ads can prove a good investment for just that kind of visitors.

B. Predict visitor engagement and use it for campaign optimisation or channel optimisation.
For example by landing page differentiation on geographical region, showing those products that have the best fit with the regional interests, taste and culture. Another example of visitor engagement differentiation is to identify and predict channel preference or timing of communication. The latter can make a huge difference for usage in social media.

C. Predict product popularity and use it for product placements.
Product recommendations are a well established technique to personalise a web page, app, or email. By using customer attributes, browsing behaviour, or situational context one can recommend products that are perceived as relevant. Companies like Spotify and Netflix exist mainly because they are able to operate recommendations effectively.

How does it work?

Practicing MLOps means that you advocate for automation and monitoring at all steps of ML system construction, including integration, testing, releasing, deployment and infrastructure management.

MLOPS

The steps include:

  • a data pipeline providing up-to-date, validated and preprocessed data (Feature Store)
  • an development environment to experiment with building models
  • a training pipeline that allows you to create ML models from the data
  • a continuous training pipeline that automates created training pipelines
  • a model deployment process that allows you to take your (chosen) model to production
  • a model performance monitoring process that (automatically) checks if model performance is still good or decaying.

Where to start?

It doesn’t start with data or technology. It does start though with a business plan of how to monetise a generalising model by defining the use cases that serve the strategy.

Next you need a multidisciplinary team. The team usually includes data scientists or ML researchers, who focus on exploratory data analysis, model development, and experimentation. They build a training pipeline delivering models, test them and try to convince the business of their added value. Next to that, experienced software engineers are needed to build production-class services that deploy models, monitor model performance and can automatically retrain them. Finally, operations professionals should take responsibility for monitoring the overall MLOps continuity.

What can Crystalloids offer?

Well, we are experienced at numerous clients in designing, implementing the entire cycle using several technologies. If you want to act quickly and don’t want to make the mistakes that others did, contact us.

CONTACT US


ABOUT CRYSTALLOIDS

Crystalloids helps companies improve their customer experiences and build marketing technology. Founded in 2006 in the Netherlands, Crystalloids builds crystal-clear solutions that turn customer data into information and knowledge into wisdom. As a leading Google Cloud Partner, Crystalloids combines experience in software development, data science, and marketing, making them one of a kind IT company. Using the Agile approach Crystalloids ensures that use cases show immediate value to their clients and frees their time to focus on decision making and less on programming.

Topics:MLOpsMachine Learning (ML)

Comments