Meet the Winners of UmojaHack Africa 2022: African Snake Antivenom Binding Challenge (Advanced)
Meet the winners · 21 Apr 2022, 12:06

A conversation with Azer Ksouri, Mokhtar Mami, and Daniel Bruintjies on their different strategies and winning solutions in African Snake Antivenom Binding Challenge (ADVANCED) was held by Zindi as part of UmojaHack Africa 2022.

1st Winner: Azer Ksouri (ASSAZZIN) from Tunisia, SUP’COM

Please introduce yourself?

My name is Azer Ksouri, I am an Engineering student at SUPCOM Tunisia. I have been practicing machine learning and data science for 2 years.

How did you prepare for UmojaHack Africa 2022?

I have been preparing by participating in many Zindi Competitions - This helped me understand how to approach different challenges and come up with different solutions to them.

Please explain your solution and what set your winning solution apart from others.

I took some time to understand the starter notebook provided by InstaDeep. I then turned the starter notebook into modular form; changing the architecture of the model - which helped in getting results faster.

What do you like about Zindi?

Zindi is a good place to start accelerating your career as a data practitioner. Zindi also supports inter-community learning and growth, where different people from different communities are able to post solutions that we can learn from. I have also sharpened my coding skills through Zindi. The competition data is not huge (in terms of volume), hence motivating new competitors and learners to indulge in various hackathons.

Words of encouragement for others, or advice that has helped you?

In any hackathon, start by going through the starter notebook, understand the problem, and look for similar problems. This will help you ideate solutions in reference to the current problem outlined in the hackathon. For beginners, the learning process takes time - be consistent and make use of online resources.

You can find a link to the notebook here.

2nd Winner: Mokhtar Mami (mo5mami) from Tunisia, INSAT

Please introduce yourself?

My name is Mokhtar Mami. I am pursuing an Engineering Degree at INSAT specializing in data science and machine learning.

How did you prepare for UHA22?

I took the time to look at past competitions and respective solutions on Zindi.

Please explain your solution and what set your winning solution apart from others.

  • For my validation strategy, I first split the signal in intervals (step=0.1) and split my data according to these intervals (stratification). I also tried grouping by ProteinId (Stratifiedgroupkfold) (it seemed logical) but my CV and LB decreased so I dropped it.
  • I tried introducing 1D CNNs after the embedding layers. This approach didn't improve the performance, but I think it helps when blending.
  • I made the architecture bigger (larger embeddings, more LSTMs ...)
  • I didn't use only MSE loss, I also used Huber Loss with different deltas. With Huber Loss, we change the loss from MSE to other losses when the loss gets bigger than delta. This makes the model more robust to outliers and gives it more freedom to experiment more (double edged) since it will be penalized less.
  • I used protbert instead of the embedding layers (for feature extraction), but it was slow and I couldn't train for many epochs so I didn't use it.

What do you like about Zindi?

It provides opportunities to African students and mainly motivates people to learn through competitions. It also offers a conducive space to connect with fellow data scientists.

Words of encouragement for others, or advice that has helped you?

For beginners, this is a process that takes time. I would advise you to look for resources and also read other people’s work - Theory is important but it also takes time to put theory into practice.

You can find a link to the notebook here.

3rd Winner: Daniel Bruintjies (DanielBruintjies) from South Africa

Please introduce yourself?

My name is Daniel Bruintjies. I started learning machine learning and data science early last year. I enrolled in a 6-month bootcamp at Explore Data Science Academy and my background is in solving problems. I also did a 2-year course in Mechanical Engineering. I am intrigued by solving problems that cut across different fields.

How did you prepare for UHA22?

I first chose a competition which I could learn the most from - I also took time to understand the starter notebook and working in different competitions also gave me the confidence and skills for this competition.

Please explain your solution and what set your winning solution apart from others.

I just worked on the starter notebook since it performed really well on the leaderboard and as I tried to understand it, I could actually get to understand which variables are highly likely to affect the machine learning model that I trained using a neural network. I split my data and trained them simultaneously on different notebooks and then ensembled the results which improved the accuracy of the model. I also added LSTM layers so that I could get more information from the amino-acids sequence which also improved the model accuracy.

What do you like about Zindi?

I like Zindi because it is mainly a platform for Africa - when working on challenges hosted on Zindi - you actually feel like you are improving the development of Africa.

Words of encouragement for others, or advice that has helped you?

Clearly defining a problem before getting started with any task is a very important skill. Exploring different competitions and different libraries to use on different tasks gives you an added advantage. Practice by doing!