I am writing this article at a beautiful late hour, a time when I find all my inspiration …
Often described as the “sexiest job in the world” or in the universe, whatever i hate this term. I tried to say that I was a data scientist to women one day and try to explain how linear regression works, and that is the opposite of sexy.
In short, the profession of Data Scientist continues to change. I will explain why from my experience of working with data scientists and especially to have been around that the reign of sexy data scientists is coming to an end.
Auto-ML and API’s
Evil tongues will say that we always need data science in our projects. Yes, exactly but remember to distinguish between data scientist (the job) and the domain.
Auto-ML or automated machine learning makes it possible to build increasingly reliable models without reinventing the wheel. Because like many data scientists, many models that are viewed and reviewed are becoming more and more accessible thanks to the rise of the Cloud.
Thus, Azure or Google cloud for example, make it possible to carry out ML without even coding and deploying more easily. Because what is problematic is the deployment of the model which represents 80% of the work. Working directly on a cloud and its APIs makes it easier to develop and put a model into production.
Which brings us to the next point …
The deployment and scalability of the model
To present an AI model, it’s cute, notebooks, anaconda, the little teddy bear … But companies want to take it up a notch. Why? coronavirus has also accelerated business transformation. Likewise, the data scientist has often been associated with R&D for years, yet we are almost in 2021, and the ecosystem is becoming more and more mature. Hence the need for the integration and production of large-scale models. And in these last points, data science enters a new external paradigm different from the old notebook locally: pipelines, versioning of models, re-training, dockerization, API calls, monitoring, automatic comparisons or performance manual… In short, a good deal of devops and data engineering.
The advent of Machine learning Engineer
And that’s another nice buzz word, but that’s what companies are looking for, because what brings value is the deployment and the entire management of the model’s value chain. He incorporates in his approach… drum roll of another buzz word…. MLOPS is sort of devops applied to the management of the model.
Too many data scientists
After talking to several recruiters as well as rather technical people, a similar pattern always emerges. “We don’t need a data scientist.” And it’s often said raw.
Another company, where I had applied as a Data engineer, the firm, clearly told me, you see them over there, they are killers, especially do not put forward your skills in data science in addition to your data engineering skills but only your data engineering skills.
Another one, “we don’t need a datascientist, there are too many of them”, the only data scientists they keep under the elbow are the tops of the top, the gurus …
Another this time on linkedin, “I get disgusted when I see the term data scientist in the profil”, so the excess of data scientists can make some people nauseous …
After being in IT Consulting company, finding a mission as a data scientist was like winning an award. Indeed, for one data scientist in a company, we need several developers, data engineers, devops …
We still need some data scientists
The data scientist is still in demand, for research, or for very specific needs. But of course, we are talking about a data scientist with a very good level, researcher…
What should our data scientist friends do to stand out?
Stop being a data scientist at first.
Specialize, either as a machine learning engineer, AI engineer, computer vision engineer … Whatever happens, new skills must also be mastered, especially techno such as the cloud, docker, and a little devOps.
There is also the way to move towards Data engineering, but the level in code, infrastructure … is greater. Finally, the ex data scientist must do more software engineering than usual than just pure math and stats …
So, add some engineering!
Bye...
