From Business Intelligence to Data Science & Machine LearningChronicling my transition from the field of Business Intelligence to Data Science & Machine LearningRajkumar KaliyaperumalBlockedUnblockFollowFollowingMar 16I’m a Manager of Data Science & Machine Learning, specialising in design & building of solutions powered by DS & ML for enterprises for their analytics requirements.
I have more than 5 years of extensive experience now in the field and like many, I did not begin my career in Data Science & Machine Learning.
I was a Business Intelligence professional with close to 10 years of experience when I decided to take the plunge into Data Science.
I’d like to share my experience of transitioning into the field in the hope that it might help a few who’re are looking to transition into the field.
It’s definitely not this hardI’d imagine it to be something like thisFor those of you who are not familiar with Business Intelligence(BI), it is a branch of technology that enabled the organization to gather all the historical enterprise data that is available to them and start using them to make business decisions both tactical & strategic.
As the field of data science & machine learning was just beginning to take off in the early part of the decade (Thanks to the article “Data Scientist: The Sexiest Job of the 21st Century” https://hbr.
org/2012/10/data-scientist-the-sexiest-job-of-the-21st-century that got published in HBR) I felt a strong urge to ride the wave.
But later felt that this is a transition that I had to make sooner or later because it was, in my opinion, a logical extension to the field of BI.
While the field of Data Science & Machine Learning was about giving business the edge by providing actionable insights and intelligence about the future (some may argue that it’s also about the present), the field of BI did the same, but using data from the past using lesser sophisticated techniques through dashboarding and visualization techniques.
Or Maybe, little harderThat said, here a few pointers that I’d like to share with all the wannabe Data Science/Machine Learning professionalsAlmost every Data Scientist is a Citizen Data ScientistGiven the sudden surge in the demand for data scientists & machine learning professionals when the industry does not have enough talent, most of the demand is being met with people who transitioned to ML & DS from fields as diverse as Bach of Arts, Psychology etc.
So don’t ever feel that you are late in the game.
In fact, if you are from the field of BI, you can leverage a lot of your existing data skills in DS/ML (Pandas Dataframe operations could very well be done on a relational table using plain SQL).
Apparently even during the age of Darwin, many scientists were ordinary people who took to science out of curiosity & interest and were hence called citizen scientists.
Hence, you could be a citizen data scientist too if you have the interest and drive in you.
2.
Have a structured learning pathMOOCs on DS & ML are plenty these days.
I chose to certify in a couple of courses.
Its okay to be slow, but be steadyMachine Learning Specialisation from University of Washington (Coursera) https://www.
coursera.
org/specializations/machine-learningThis is 4-course specialization by Carlos Guestrin and Emily Fox who founded Turi that later got acquired by Amazon, covering all important areas of machine learning — Classification, Regression & Clustering using python which is one of the reasons I chose this program2.
Deeplearning.
ai(Coursera) https://www.
deeplearning.
ai/deep-learning-specialization/This is a 5-part specialization in Deep learning offered by Andew Ng himself explaining the concepts of popular topics — Deep Neural Networks, CNN, Sequence Models et al3.
Udacity Nanodegree in Deep Learning Specialization https://in.
udacity.
com/course/deep-learning-nanodegree–nd101This is a 4-month specialization in Deep Learning concepts using Pytorch which is one of the emerging and widely used DL frameworks in the deep learning research community today.
I chose this primarily because I wanted to stick to a framework with Python flavour for easier adaptability.
Btw, this is not an exhaustive account of courses that I tried and there are many more courses that I’ve done partially but not naming them here.
I’ll cover all my course experience in an exclusive story later.
3.
Get ready to be hands-onBe ready to code, regardless of your titleWhile these courses help you to get a grounding on the concepts with some hands-on projects, it's important that one practices these concepts in order to build expertise.
One of the activities that helped me was to participate in competitions in the popular competition sites such as Kaggle, Analytics Vidhya to keep in touch with your learning.
For eg.
I took part in the competition “Loan Prediction Challenge III” from Analytics Vidhya and secured 31st rank in the overall leader board.
https://datahack.
analyticsvidhya.
com/contest/practice-problem-loan-prediction-iii/lbEven though this is a playground competition, it helps you to build confidence in the skills that one has learnt.
4.
Open other channels of LearningCourses are one of the sources of learning, but it cannot be the only source of learning.
It's important to tap into other sources of learning as well in order to develop a wide and broad perspective of topics.
Here are other learning activities that I undertake.
a) Other online learning content — Subscribed to “Practical Deep Learning with PyTorch” by an NUS researcher on Udemy.
https://www.
udemy.
com/practical-deep-learning-with-pytorch/learn/v4/content”I liked the way he explains the concept of a convolution filter as a filter that looks for specific shapes/patterns in an image.
This gives a different perspective on the concepts learnt in the courses mentioned above.
b) Follow the works of topmost influencers in the field — I follow Jeremy Howard@jeremyphoward, Rachel Thomas@math_rachel (co-founders of fast.
ai an advanced pytorch based framework that makes Deeplearning coding a lot easier), Andrej KarpathyAndrej Karpathy,Andrew Ng@AndrewYNg , Jason Brownlee@TeachTheMachine ,Soumith Chintala@soumithchintala You may follow them on twitter or read their blogs to keep a tab on their works from time to time.
Since I chose python based frameworks for my learning, almost all of the above use Pytorch as a framework or a python based libraries for ML & DS.
So my list could be biased that way, but essentially the point is that you’ll need to follow those people who are the biggest influencers in the field to be in tune with what they are doing.
c) Read blogs — Follow blogs of top technology companies that do path-breaking work in ML & DS.
I choose the following companies blogs closely through Feedly.
Facebook AI Research (FAIR), fast.
ai, Google AI blog, AWS ML Blog, Open AI blog.
This is not an exhaustive list by any account, but I’m just calling the top ones5.
Don’t wait to land the “perfect” ML/DL projectOnce the learning is done or even when the learning is on, try to implement the learnings in your current project by proposing ML, DL concepts in your solution design.
From my experience working with clients from Fortune 500, most clients don’t specifically ask for an ML/DL project.
Their priorities are always meeting the business objective and the design & approach of the solution is entirely left to us.
So, I advise you to look for opportunities where you can put some of the things you learnt to practical use.
For instance, I took up implementing text classification tasks using ML and simple sentiment classifier using DL in my projects.
6.
Take up pet projectsIn addition to projects for clients, taking up pet projects is another great learning opportunity for you and also the organization.
This is especially true in the case of DL as most of the projects are in the realm of unstructured data in the form of text & images.
So working on some pet project could not only help your knowledge grow but also gets your organization to take notice of the skills that you bring to the table and for your team to get hooked onto these concepts.
For instance, I’d taken up the task of building a prototype to automatically detect license plates of cars using object detection technique.
More on this later, but below is a snapshot of tool that my team & I built.
License Plate Detection.