Why do both matter?
More and more organizations – small and large – are coming out of the Proof of Concept phase of developing machine learning models. They’re now up to the task of putting these machine learning algorithms into operation and let them create the promised value. We call this MLOps, or machine learning operations.
To be able to get machine learning algorithms to work and maintain it, organizations need DevOps/IT capacity and tools to put their algorithms into production.
Script or UI?
One of the big questions in building machine learning operations is to whether work from the command line or use different approaches, such as a User Interface (UI). Both alternatives have advantages and disadvantages. Luckily, with UbiOps you can choose how you want to work. Besides intuitive UI and our Python client library, we also released a CLI to interact with the UbiOps API from the terminal.
Advantages and disadvantages of UI and CLI
Whereas a User Interface is often much more user-friendly, it gives less flexibility. On the other hand, scripting from the command line can make things nontransparent, whereby you easily can lose control over your machine learning operations.
The question to rather script or to use a UI will be answered differently by data scientists and IT/DevOps engineers. Data scientists have the preference to build machine learning algorithms, not deploying it. A User Interface in machine learning applications can help them to make the process of deployment much more intuitive and user- friendly. They don’t need specific DevOps knowledge to put machine learning into operation.
DevOps/IT engineers are specialized in building machine learning operations. When a machine learning model needs to be deployed, they need to make machine learning operations reproductive and sustainable. A UI can give them less control of the flexibility. For example, when a data scientist requests to upload 10 versions of the same algorithm, the IT/DevOps engineer will prefer to write a script from the command line rather than upload it 10 times in the same manner on a UI.
Moreover, it also depends on whether a User Interface can be used in a specific environment or infrastructure. Many organizations might prefer to install a virtual private cloud (OnPrem). Another preference is to use their own servers to let their machine learning operations run on top of. Scripting might be the way to go. A User Interface might not be feasible if it needs an outside web-connection.
Customer Case with UbiOps
“Building such an infrastructure ourselves would cost us at least half a year, while we want to quickly bring value to our clients. With UbiOps we can win highly competitive tenders because we can develop new functionalities and deploy applications within a matter of weeks!”
Gradyent, a company active in energy networks, has decided using UbiOps for operationalizing and maintaining their machine learning algorithms on large scale. One of the reasons why Gradyent uses UbiOps is to be able to script against and integrate with the platform and use the UI for its data scientists in an OnPrem secure environment.
UbiOps gives a fundamental basis to let their machine learning operations run, without compromising on using a UI or to script from the command line. Gradyent uses the client libraries for its customers as well as the API that UbiOps provides to communicate with. UbiOps and its UI can run anywhere, on public/private cloud, on your own server or even laptop.
Conclusions
As this article explains data scientists and DevOps/IT engineers can have different preferences but need to work together effectively. Collaboration is one of the key challenges for organizations and needs a lot more than only tooling. Interested in this topic? Check our article on supporting a happy marriage between the data science and software team.
Do you want to stay up to date with similar topics? Follow us on LinkedIn!