Welcome, avid readers and tech enthusiasts, to the thrilling culmination of our much-anticipated machine learning hackathon! After weeks of intense competition, countless lines of code, and innovative problem-solving, we are thrilled to unveil the brilliant minds behind the top three winning submissions.
First Place;
Introduce yourself;
My name is Katende Brian. I am a third year software student of Software Engineering at Makerere University with a specialized interest and enthusiasm in Machine Learning and Artificial Intelligence as am a practicing Machine Learning Engineer.
How did you prepare for the hackathon?
The preparation was quite easy as I have participated in similar challenges concerning Natural Language Processing. Though each challenge comes with its own problems, in Machine Learning the success is research about the particular problem you are facing.
What was unique about your solution?
To begin with the challenge had limited data and this posed a challenge for training of the models. I decided to come up with an ensemble model consisting of Logistic Regression and Random Forest. Logistic regression really works well on small datasets while Recurrent Neural Networks are the best on large datasets. Pretrained models in this case did not help me at all but rather getting more random data for my models to learn from.
Any last words?
I would like to thank Pollicy Uganda for such an initiative to put out our skills and the resources invested in this competition and would encourage the organizers to set up more of these hackathons solving different world problems.
Second Place;
Introduce yourself;
I’m Kisejjere Rashid, a dedicated self-taught data scientist with 2 years of hands-on experience. I am currently also still pursuing a Bachelor of Science in software engineering at Makerere University in my 3rd year. Excited to be part of my first official competition, I extend my gratitude to the host for organizing this incredible event that challenged me to elevate my skills.
How did you prepare for the hackathon?
I carried out a lot of research on different efficient and effective ways of fine-tuning models by reading a couple of different papers.
What was unique about your solution?
For my solution, I opted to train the robust Roberta large transformer model. Renowned for surpassing its counterparts like BERT, ALBERT, and XLNET, it excels in processing input sentences with self-attention and generating contextualized word representations. The model was trained on a substantial and superior dataset, ensuring state-of-the-art performance.
To accelerate the training process, I harnessed the power of Tensor Processing Units (TPUs). This not only facilitated swift training but also provided the flexibility to retrain the model and modify hyperparameters without concerns about prolonged training times.
To enhance the model’s performance, I implemented early stopping callbacks to prevent overfitting. Additionally, I employed the model checkpoint callback to save only the best models after each epoch. The inclusion of layer normalization, inspired by a referenced research paper, further elevated the overall accuracy of the model, contributing to the success of my solution.
Third Place;
Introduce yourself;
I am Bwenge Kyangwi josue, currently pursuing a Bachelor of Science in Artificial Intelligence and Machine Learning at ISBAT University. My academic journey reflects my passion for the dynamic fields of data science, mathematics, and robotics. Eager to apply my knowledge and skills, I actively engage in competitions on Zindi Africa under the username Bwenge840. These experiences not only enhance my practical understanding but also fuel my enthusiasm for contributing to advancements in these exciting domains.
How did you prepare for the hackathon?
My preparation for the hackathon stemmed from a continuous commitment to personal and academic growth. As a university student, I have actively pursued opportunities for learning by engaging in online classes and participating in various competitions. Competitions, in particular, have proven to be invaluable experiences, offering me practical knowledge that I can directly apply to benefit my community.
These events have not only equipped me with technical skills but have also instilled a sense of adaptability and problem-solving, qualities that I believe are essential in making a positive impact on society.
What was unique about your solution?
In the “Data Ladies End of the Year Hackathon,” my solo effort was pivotal in crafting a winning solution. I initiated the process by delving deep into the dataset, ensuring a comprehensive understanding of the challenges posed by the hackathon’s theme. Leveraging advanced natural language processing techniques, particularly with the integration of state-of-the-art pre-trained models such as BERT, I aimed to extract nuanced insights from the provided text data, enhancing the robustness and accuracy of my solution.
What truly distinguished my approach was the combination of technical innovation and meticulous code organization. Despite working alone, I maintained a well-documented and modular codebase, prioritizing clarity and efficiency. This not only facilitated the development process but also ensured that my solution was not only technically sound but also accessible and understandable to others.
Conclusion
From decoding the secrets of clickbait alchemy to crafting models that redefine the game, each winner has left an indelible mark on the landscape of text prediction. The fusion of creativity, technical prowess, and relentless determination has not only raised the bar but has set a new standard for future hackathons. This journey doesn't end here, visit our GitHub to explore the code, unravel the complexities, and perhaps uncover inspiration for your next data-driven adventure. Check out the full final leaderboard here!
