Where does UbiOps fit in the data science life cycle? The importance of deployment & serving.

If your company is doing anything with data science and analytics, you will probably recognize that there are multiple stages to a project and that there need to be many different teams and skillsets involved.

Table of contents

1. Data science lifecycle
2. The importance of serving & orchestration
3. Other stages of the data science lifecycle
4. Integrations & AIIA
5. About UbiOps

On a high level, we can distinguish the experiment phase and the operations phase. Usually, these phases require different people and different capabilities within your organization. One of the biggest steps to make is the move from the experimental phase to running an algorithm in production. The skill sets involved and the required technologies are very different. Of course, running in production can have different meanings, depending on your organization and the processes already in place. A start-up developing a proof of concept for an app to recognize food items has very different needs than an utility company running demand forecasting models where the output affects the operation of a power grid. However, no matter how large or small your project is, a positive return on investment (ROI) is only possible if you run your analytics on live data and give end users a taste of the actual application. For this, you need a serving infrastructure for your models and developing one doesn’t have to do anything with data science, it’s software. Many smaller teams often start with running a Dockerized version of their model with a web server on a cloud instance. Or they pick up an open source tool like Kubeflow. Actually, there are a ton of ways to get to a working setup, but only very few will ‘serve’ you well. We see many teams start out building a solution, but get stuck further down the road. The Docker & cloud instance approach doesn’t turn out to be very flexible and any change to the code and change in load requires manual work. For many open source frameworks, the part after the ‘hello_world()’ becomes a months-long trip down-the-rabbit-hole.

The importance of serving & orchestration

We at UbiOps cover exactly the part that you need when you’re done experimenting and want to run your models on live data. We aim to make this as easy as possible for data scientists and empower them to create value and a positive return on investment quickly with their work. There are many things that come into play when it comes to model serving. Especially at scale. To guarantee availability and uptime, a serving infrastructure needs redundancy. Models often require different code dependencies and software packages, which requires them to be packaged in individual containers to isolate their runtime environments. Also, different models can have different scaling and compute resource needs and benefit from individual horizontal scaling. More importantly, all of this needs to be secure and must be monitored well. Also, the serving infrastructure needs to be made for efficient updating of models as machine learning has more frequent release cycles than regular software applications.

Other stages of the life cycle

Of course, the journey often doesn’t end here, but deployment & serving is an important link in the chain. If your team becomes more advanced, you likely want to check out options for monitoring and explainability, data lineage, automatic retraining or advanced deployment techniques too.

Integrations & AIIA

The space of tools for covering the data science lifecycle is growing like crazy. Machine Learning Operations, or MLOps, now is a central theme in many organizations and its importance will only grow. There is a growing number of amazing tools out there to help you out. Not only cloud provider services, but companies with specialized offerings, together creating a new canonical stack of MLOps software solutions. The AI Infrastructure Alliance proves that there is a growing solution space of specialized tools, integrating with each other to provide the market with a state-of-the-art stack covering the data science lifecycle and MLOps. We as UbiOps are working hard to provide you with a managed deployment and serving platform to get your models up and running in no-time. But we also recognize that different teams have different needs that you cannot solve with an all-in-one solution. Therefore we focus on integrations with others and providing universal interfaces. Together we can improve the way data science is implemented and used.

About UbiOps

UbiOps is an easy-to-use deployment and serving platform built on Kubernetes. It turns your Python & R models and scripts into web services, allowing you to use them from anywhere at any time. You can embed them in your own applications, website or data infrastructure. Without having to worry about security, reliability or scalability. UbiOps takes care of this for you and helps you get ahead with MLOps in your team. UbiOps is built to be as flexible and adaptive as you need it to be for running your code, without compromising on performance or security. We’ve designed it to fit like a building block on top of your existing stack, rather than having to make an effort to learn and adopt new technologies. It lets you manage your models from one centralized location, offering flexibility and governance. You can find more technical information in our documentation -> www.ubiops.com/docs  To help you getting up to speed with UbiOps, we have prepared some examples and quickstart tutorials -> www.ubiops.com/docs/tutorials/quickstart/ If you have any stories on your journey from experiment to production with AI, let us know.