Microsoft Machine Learning Hackathon Results in May


    At the end of May 2016, we successfully held the third machine learning hackathon. The topic of advanced data analysis in recent years is gaining more and more popularity, and we are pleased to contribute to its development. For two days we gathered more than 150 people under the roof of our office, who, during a short acquaintance with each other, quickly chose captains and divided into 24 teams.

    Hackathon participants were offered a choice of solving 4 data modeling tasks and 1 visualization task using Microsoft technologies: Azure Machine Learning machine learning service and Power BI business intelligence service.

    Customers whose business tasks the participants had to solve for two days were:
    • "Magnitogorsk Iron and Steel Works" (determine the silicon content in the furnace)
    • Community of Entrepreneurs “Business Youth” (classify customers as hot and cold)
    • Network of clinics "Mother and Child" (to predict the sale of groups of services)
    • And one of the largest Russian banks, which took part anonymously. (estimate future customer turnover on transactions)

    More details about the tasks can be found on the event website .

    At the end of the first day, we determined the best in the data visualization nomination, the winner was the Small Data team, which received a Nokia Lumia 650 phone as a prize. The team made several beautiful and representative dashboards and also used an unusual technique for clustering customers.

    Congratulations to team captain Sergei Chernov and the team itself with a victory!

    Other teams also demonstrated interesting options for interpreting data:

    At the end of the second day, the modeling decisions were defended before the jury.

    The winners in the task of the company “Mother and Child” were the ZeroBit team (RMSE = 1,325,097.9).

    The best solution for the “Business Youth” task was demonstrated by the “5 seconds to win” team (AUC = 0.634199).

    Since no team showed satisfactory results for business customers on the tasks of the Bank and the Magnitogorsk Iron and Steel Works, it was decided to extend the Hackathon until May 29 and allow participants to continue to compete for prizes remotely. According to the results announced at the beginning of June, the best result for solving the MMK task remained with the Catalysis team (captain Alexander Orlov), and according to the results of solving the bank's task, the Home CR team (captain Mikhail Shabalin) took the lead. Nevertheless, the prize to the team solving the MMK problem was not given, since the result obtained did not meet the accuracy criterion set by the customer.

    At the moment, we continue to communicate with the captains of these teams and customers in order to maximize the quality of the results.

    Invited experts who demonstrated a high level of knowledge of machine learning technologies provided valuable assistance in conducting the event and evaluating solutions.

    The experts at the event were:
    • Alexander Fonarev, Alexander Kuzmin and Mikhail Gavrikov from Rubbles, a member of the SBDA Group. Company Rubbles specializes in the application of machine learning methods and analysis for a variety of business tasks - for example, to personalize the marketing and banking services, building intelligent interactive chat robot platforms or predictions of demand in
      major stores. Now Rubbles works with the largest financial
      and technology companies (Sberbank, Mastercard, Alfa-Bank,
      Qiwi, HSBC and others).
    • Oleg Kharatov from the IT consulting company Manzana Group, Microsoft Certified Gold Partner.
    • And also: Alexey Peshekhonov, Evgeny Grigorenko and Dmitry Petukhov ( @codezombie ) .

    We thank everyone who has been with us these two days, we will be glad to receive feedback,
    and we are waiting for everyone at the new hackathons!

    Also popular now: