Leveraging ML Ops to Enhance Your Data Science Factory
Machine learning falls into a category of technology currently experiencing hyper-exponential growth as enterprises capitalize on its ability to transform data into insightful action. Like any hyped technology, machine learning is not without current limitations; however, companies operating on the cutting-edge are finding innovative ways to integrate machine learning into an impressive bottom-line. And it’s no longer just pet projects for elite Fortune 500 brands—everyone is joining the fun.
With this rapid evolution, the role of data professionals is being reconceptualized; leaders increasingly understand data-related infrastructure in terms of the factory model, giving rise to the notion of a data science factory. Data goes in, actionable insights come out, and everything happening in-between falls into the “making the sausage” category—nobody wants to know too much.
On one end of the factory, you have the dizzying collection of platforms to manipulate data (Python, H2O, TensorFlow, R, Scikit-Learn, Keras, SAS®, Openface, Caffe2, Watson, Google, Azure, AWS ML cloud APIs, and the list goes on). On the other end of the factory, you have the final outcomes: recommendation engines, IoT programming, bots, or any type of automated decision. The most critical element of this factory model is the feedback loop existing between the production and the results; this relationship creates the “learning” capacity in machine learning.
We’ve discovered a critical flaw in most companies' production methods. Data scientists may design promising new models in two weeks, but it takes the better portion of a year to actually integrate these models into business practice. The result is a bulky, expensive, and frustrating process culminating in a perpetual technical debt caused by quick-fixes and inefficiencies. Based on this approach, the term “data science factory” seems more like an ideal than a reality.
Most machine learning production loses footing somewhere between model production and deployment, with no single person or software to blame. Coordinating the efforts of dozens of platforms with final user environments (websites, apps, proprietary software) results in complex and cumbersome processes. If data science factories followed Henry Ford’s model, the smallest team would have no fewer than 20 people, with each contributing narrow expertise. As most CTOs can attest, that’s not an acceptable solution.
Recognizing this inability for teams to achieve streamlined production, we decided to design an alternative solution. By providing data scientists a comprehensive production environment, we could drastically reduce the time between model development and deployment from months to days—the financial impact would be radical, followed closely by the impact on the data team’s sanity. In order to achieve this, data scientists and engineers would need an integrated solution providing a cohesive fabric between the popular platforms and the final interaction points. The environment would need to allow data and machine learning engineers to test and score models while simultaneously looping feedback from the interaction points; it would have to serve as a repository, performance monitoring center, and allow simple model management. Finally, the entire functionality would need to be captured in a modern, intuitive dashboard, to help bridge the persistent communications gap between data teams and partnering executives or clients.
The resulting platform surpassed our expectations. With features like a decision studio and seamless end-to-end integration, Quickpath’s partners are realizing factory-like data science production that would bring a tear to Mr. Ford’s eye. Many of the individual components already existed, and the teams had heaps of skill and talent, but without the woven fabric connecting it all, they couldn’t achieve strong results.
Data science, as a field, did not evolve from a singular practice, but a natural cohesion of multiple disciplines with strong roots in statistics. Machine learning, engineering, big data, data mining, data analysis, and database management are all in orbit around rapidly-advancing technology, which creates an inevitable degree of chaos. Any company striving to integrate machine learning faces this reality. Still, some leaders are leaning into more advanced ML Ops methods, working with partners like Quickpath, to break their team into the next realm of possibility.
Organizations are realizing that effective departments don’t function in chaos—organized chaos, maybe, but not plain chaos—and that data science is no different. It’s no secret that most corporate AI/ML projects never live beyond the test phase, and it’s not a coincidence that ML Ops began gaining traction in 2018. In order to successfully navigate the waters in a nascent and complex field, operations are perhaps the most integral ingredient.
It’s not unusual to hear executives discuss future-proofing a company, anticipating the next unexpected turn. Some leaders even pursue AI/ML projects to this end, treating data projects like corporate storm cellars, separate from the bottom line. However, the rate of change is accelerating, and the once-distant future of data-driven enterprises is now firmly a reality for many leading brands.