Since the dawn of time and up until 2005, humans generated roughly 130 exabytes of data. By 2015, we generated 7,900 exabytes of data. This figure is estimated to be around 40,000 Exabytes by 2020! It doesn’t take someone special to conclude that, with the inevitable growth of computational power, data generated by humans is rising exponentially. This should imply that Machine Learning is the future.
A lot of organizations, providing technical solutions or not, have realized the potential of Machine Learning and this has led to a surge in the number of jobs and applicants related to Data Science. Hiring challenges faced by applicants are becoming increasingly tedious, where in addition to the time restrictions, applicants are expected to solve a problem by developing an entire end-to-end technical approach.
In classes, students build on what they’re taught and are often guided on how to solve problems. Facing such hiring challenges requires the students to develop a different approach, where, whilst keeping the bigger picture in mind, every little detail needs their attention too. To help students build their confidence before they face such challenges, GreyAtom recently hosted a Hackathon for a duration of 2 days, starting on 2nd December 2017.
This was an ‘internal hackathon’ which means that all the participants were students at GreyAtom. Teams were segregated systematically using “Commit.Live”, a state-of-the-art online educational platform designed and built by GreyAtom for its students. The performance of every student, which encompasses code-efficiency and a lot of different parameters, is measured throughout the duration of the course through “Commit.Live”. Thus, for the sake of simplicity, these scores were used to build the teams where the average performance score of each team was approximately the same.
In order to expose participants to such an intense ecosystem, we decided to make each team pick 1 of the 4 FinTech Problem Statements presented to them, which are listed below.
- Build a model to approve/ disapprove a loan for a prospective applicant based on his/her profile.
- Predict if a client will subscribe to a direct marketing campaign for a banking institution.
- Build a model to compute the probability of default for a Taiwanese Credit Card Client.
- Predict whether a particular company will default within 5 years, given its financial statement data.
Traditionally, Hackathons call for teams to solve one specific problem who are then judged on a specific metric, for example, Root Mean Squared Logarithmic Error and other factors like efficiency and implementation. However, for the purpose of learning, participants had the flexibility to choose any problem statement which motivated them to stay focused for a duration of 2 days. However, each team’s code and presentation was judged by a panel of 2 members and by other teams as well. A team would receive a maximum of 50 points based on the following criteria:
- Documentation + Presentation
- Model Building + Parameter Tuning
- Exploratory Analysis
The winning team chose to work on 4th problem statement mentioned above and were successful in building a model that gave an AUC (Area Under the Curve – more specifically: Area under the Receiver Operating Characteristic Curve) score of around 0.9. On a small test set, their model correctly predicted the outcome of the company i.e. whether it went bankrupt or not.
Hackathons prove to be tremendous learning experiences for participants. I believe that it’s not really easy to solve a Machine Learning problem in just 48 hours. It challenges you in many different ways. If you’re solving any sort of problem out there in the real world using Machine Learning techniques, you’ll realize that it is your team that determines your success.
Our students who spent the weekend competing at this particular Hackathon, now, have an idea on how to prepare themselves, mentally and physically, for more competitive Hackathons in the future and for crucial hiring challenges as well. Saiprasad Balasubramanian, one of the participants, said, “This was indeed a great learning experience! My team didn’t win, but I guess that wasn’t the motive of this event. I got to experience how to work collaboratively as a part of a team, especially in a chaotic environment. This hackathon was much more than just picking some algorithm to solve a particular problem. I actually made a list of my learnings.” He continued to elaborate on his experience and this led us to ask most of the participants to share their experiences as well.
A lot of the participants spoke about their mistakes, which is good because you really do learn from them and not repeating them in the future can help save a lot of time and improve overall efficiency. Therefore, to help individuals who are relatively new to the whole “Hackathon” ecosystem, we listed out the things to take note of for future Hackathons.
After successfully concluding the Hackathon and after learning about the positive experiences of all participants, GreyAtom is motivated to host a number of Hackathons in the near future.