HSBC’s Akash Gupta has won over 45+ machine learning hackathons to date. The MachineHack Grandmaster has come second thrice in a row and is currently ranked sixth on the platform. “I’ve always been fascinated by numbers and patterns. I got very curious about algorithms – how they are made, how they work, and what we can do with them– after I took Andrew Ng’s machine learning course,” said Akash Gupta.
The data scientist spoke about his MachineHack journey in an exclusive interview with Analytics India Magazine.
AIM: What were the initial challenges, and how did you address them? Akash Gupta: I entered the field in 2016, and have faced many challenges initially. Coursera and EdX were the only two platforms at the time to get people like me started. First, I signed up for Andrew NG’s Machine Learning course and then followed it up with IBM’s data science professional course. Later, I enrolled for the Mathematics for Machine Learning course by Imperial College, London. After taking up a lot of courses to improve my skills, I told myself, “It’s competition time.” I have learned many things from hackathons such as:
- Understanding the problem in-depth
- How to choose the best algorithm to train my model
- How to improve the accuracy of the model
- Understanding the business impact of the model
- Ensuring positive business outcomes with the model
With every hackathon, I solved new challenges and improved my knowledge. I took one step at a time to reach the grandmaster level.
AIM: What about coding excites you the most?
Akash Gupta: After I learned to code, I realized it gave me the power to turn virtual things into reality. And I can put this ability to test at hackathons.
Hackathons give me a chance to approach problems with unique ideas that are both logical and mathematical. It works perfectly for me because I like to solve puzzles, play with numbers, write codes, automate and make recommendations and predictions.
AIM: How do you approach a problem?
Akash Gupta: My approach is very simple – always try to connect with the problem. After understanding the problem, I try solving it using the basics. Typically, this is enough to tackle most of the problems. Regardless of how good you are at building model or feature engineering, it’s pivotal from a business and logic perspective to first understand the problem.
AIM: What does your machine learning tool stack look like?
Akash Gupta: I use Google Colab or Kaggle Kernel platform to write the code from scratch. From a tool’s point of view, I use Visual Studio, H2o.ai, Weka, Jupyter (Anaconda), Google Colab, Kaggle Kernel, System GPU, MS SQL, Google Cloud Platform, Amazon Machine learning, etc. Sometimes, I use Spyder and PyCharm for ML projects. From a framework point of view, I use scikit-learn, PyTorch, TensorFlow, fast.ai, Keras, Google Cloud ML Engine, and Spark MLlib. From a project point of view, I prefer to use local systems.
AIM: How to prepare for your first hackathon?
Akash Gupta: Here are the basic skills you need:
- Ability to understand the problem
- Machine learning, modeling & data science skills.
- Hands-on practice with Pandas, NumPy, scikit-learn, and Matplotlib packages.
AIM: What’s your biggest pet peeve about hackathons?
Akash Gupta: Time series forecasting. Most of the features I built from timestamps have failed. If I used a pattern view, then it performed well. Using more features gives a good score, but my CV score tends to be low. That’s why I never used it in my final model.
AIM: What’s the worst experience you had as a coder?
Akash Gupta: My coding experience has been amazing. The only time I get irritated is when I commit silly mistakes, like forgetting a comma or space. For example, when I write a SQL Query and it’s running in my system, I get an error when I’m trying to run it on other hosted platforms. That is so irritating.
AIM: What drew you to MachineHack? Tell us about your journey so far.
Akash Gupta: MachineHack is one of India’s best hackathon organizers and data science knowledge portals. I started actively participating in ML hackathons during my fourth semester at the university. At the time, I was using many platforms like MachineHack, Heackerearth, Kaggle, etc. MachineHack comes up with a lot of hackathons regularly. On top of that, the MachineHack team is always supportive at every step of the way.
AIM: What was your first MachineHack competition like?
Akash Gupta: MachineHack’s second Hackathon was my first hackathon on their platform. I still remember it was around vehicle safety, and Intel organized it. I have enjoyed building the model for this hackathon.
AIM: How did it feel when you became a MachineHack Grandmaster?
Akash Gupta: Machinehack had a global leaderboard in the old UI. I got the title of a Grandmaster around the time the team launched a new user interface. The new UI allows users to show their MachineHack profile on their resumes. MachineHack provided everyone with a title based on rank. So when I got the Grandmaster title, I put it on my resume–it empowered me and made me feel like a strong coder.
AIM: Tips to ace Machinehack.
Akash Gupta: The only way you can get on top is by participating in as many Hackathons as possible. Solve as many problems as possible, get more knowledge and understanding of different problem statements, and join MachineHack boot camps. They will help you learn, practice, and improve the accuracy of models using new techniques and tricks.