MachineHack Grandmaster Tapas Das is a data engineer with over 9+ years of experience in building scalable, optimised, customer-centric IT solutions. The Kaggle Notebooks expert has participated in 32 MachineHack hackathons and is among the top 10 champions worldwide. 

“I can only work efficiently if there are minimal or no distractions. So, for me, getting into the zone is pretty much sitting in a corner with my laptop, headphones on, blasting rock metal at full volume. I need music to organise my thoughts and come up with better solutions,” said Tapas Das, Delivery Manager, The Math Company.

Analytics India Magazine got in touch with this alpha geek to gain insights into his data science journey and his hackathon exploits.

AIM: How did your fascination with algorithms begin?

Tapas Das: I first got interested in this subject in 2018, having come across an article about how Google Search predicts user searches so efficiently. After that, I started researching machine learning and how the different algorithms work. Then, I went through different MOOCs like the “famous” Andrew Ng ML course and the Deep Learning Specialization course on Coursera.

AIM: What were the initial challenges, and how did you address them?

Tapas Das: I had a very clear intention when starting the ML journey. I wanted to build a machine learning model from scratch without using popular frameworks like PyTorch, Tensorflow, Caffe etc.

I spent a significant amount of time learning and coding the basic ML building blocks, like perceptron, forward-propagation, backwards-propagation, gradient-descent, etc. Then, I built a basic neural network classifier for classifying dogs vs cats. Although it took two whole days to train, without any of the GPU optimisations, the final accuracy of 75% was the “Eureka” moment for me.

AIM: What about coding excites you the most?

Tapas Das: I’ve always been fascinated with different programming languages. I started coding in C and eventually upgraded myself to Python. The area that excites me the most is the handshake between the software and the hardware – how can we write more efficient code to use minimal hardware resources and run within milliseconds.

AIM: What does your ML tool stack look like?

Tapas Das: Well, that depends on the ML problem we’re trying to solve. I always start with basics like linear regression or Elastic Net if it’s a regression problem, or logistic regression or SVM if it’s a classification problem. Then I slowly upgrade to the tree models (Random Forest, XGBoost, LightGBM) or neural nets (Tensorflow or Keras).

In terms of libraries, scikit-learn is like a god’s gift and has always been my go-to library. I rely heavily on FeatureTools for feature engineering and Optuna for hyperparameters search.

AIM: How to prepare for your first hackathon?

Tapas Das: For anyone starting on the “ML hackathon” journey, my suggestion would be to start by learning EDA. The more you understand the data and the domain, the better your clarity when doing feature engineering or feature selection.

I usually prefer to perform all EDA manually instead of relying on auto-EDA libraries like SweetViz or Pandas-Profiling. This helps get better data insights by tweaking the visualisations as per need. I rely on Matplotlib and Seaborn to build the visualisation.

AIM: What’s your biggest pet peeve about hackathons?

Tapas Das: Data leakage! That drives me nuts. It makes the whole leaderboard senseless, and then everyone’s competing to optimise the 4th or 5th decimal of the metric.

AIM: What drew you to MachineHack? Tell us about your journey so far.

Tapas Das: I have been participating in different hackathons on the MachineHack platform for a while now, and I love the way the platform allows anyone, regardless of background or prior experience, to compete on a level playing field where often the only thing that matters is optimising a metric. 

Winning solutions from previous hackathons are an invaluable learning resource that I highly encourage aspiring participants to leverage. Doing a single ML hackathon teaches you more than any book or course ever could. It’s fun to compete with the greatest minds in data science.

AIM: What was your first MachineHack competition like?

Tapas Das: I’ve participated in 32 hackathons so far on Machine Hack. But it all started with the “Predicting Food Delivery Time” hackathon. Being a noob, it was a shocker participating in my first ML hackathon and failing gloriously on the leaderboard.

Afterwards, I thoroughly analysed the winner’s approach to solving that hackathon to gauge my weak points. I was blown away by the sheer uniqueness and innovative quality of each winning solution.

AIM: How did it feel when you became a Machinehack Grandmaster?

Tapas Das: When I became a Grandmaster, the feeling was similar to the last bencher in school solving a problem that even the top students failed to solve. I can’t describe it any other way. It’s a mixture of joy, ecstasy, surprise and shock.

AIM: Tips to ace MachineHack.

Tapas Das: Be very passionate about what you’re doing, stay motivated even though you’re not performing well in some hackathons, and keep learning and exploring different approaches. Sooner or later, your efforts will bear fruit.

As Eminem, rapped:

“You better lose yourself in the music, the moment

You own it, you better never let it go

You only get one shot, do not miss your chance to blow

This opportunity comes once in a lifetime yo!”