Impressions from my first Hackathon
Two weeks ago, I participated in my first hackathon, organized by ENSAE ParisTech and Microsoft Research. The event gathered 50 students for a 24-hour data science competition, based on data provided by the charity organization Croix Rouge (Red Cross). There were two categories of challenges:
Predicting the daily need at each Croix Rouge center up to 2 weeks in advance
Determining where to install new centers in France (the one my team and I chose)
It was an amazing experience, and I am very honored that my team received the first prize in its category. A picture is worth a thousand words, so here goes the winning team, aka the Data Buyers Club:
I am really proud of the work we accomplished in less than 24 hours, with little sleep and a lot of coffee/soda/pizza/etc. Generally speaking, I was very impressed by everyone’s work and the level of involvement in the competition. There were a lot of interesting models and insights, and I hope it will help Croix Rouge and the people who need it.
Here are some “impressions” from the event and our prize:
We spent 80% of our time on data loading and preparation. Machine Learning was only 20% of the job.
Big Data is hard and it requires a lot of time to handle it. Sometimes it is interesting to turn it into “Small Data”. We chose to drastically reduce the geographical scope of data, so we transformed a 10-gigabyte database into a table that could fit in Excel.
Sometimes, the simplest models are the most efficient. Most of our work was based on a linear regression. We tried Ridge and Random Forests, but it took too long.
Presentation is key. In the world of Data Science, great code is nothing without visualizations and storytelling.
To conclude, I would like to thank the team of advisers at Microsoft Research for their help during the hackathon, especially Xavier Dupré, who teaches programming at ENSAE ParisTech. His blog is an amazing resource to learn Python, algorithms, and Machine Learning.
P.S. I cannot disclose the results of our work since the data provided by the Croix Rouge is confidential. But I plan to write a summary of the methodology we used and to post it on my Github in a few weeks.
