"Programmer - Analytical thinking = 1C programmer" and other GoToHack results

From 9 to 11 December, our second hackathon GoToHack for children under 20 years old was dedicated to the use of machine learning technologies in education and HR.

Following the results, each team presented the results to the experts, the best were awarded with various prizes: from university entrance to a grant of 100,000 rubles for a trip to an international conference.
')
About projects and funny moments can be read under the cut.

We invited 90 desperately eager “typewriter” schoolchildren and students to participate, who were able to pass the qualifying online stage of the hackathon in the form of a task for Kaggle to predict the outcome of an online course based on real data from a partner - Stepik.org platform. We wrote about it in detail here .

As for the first time, the hackathon began with an intensive training and master classes:

• Alexander Panin from Yandex conducted a series of introductory master classes for beginners,
• Kamil Salakhiev from Innopolis University gave a master class on MapReduce,
• Muammar Il-Shedivat from Yandex told the guys about working with speech.

Then presentations of cases from guests of the hackathon began: Alexander Kukushkin talked about the study of Moscow schools, Alexander Petrov from E-Contenta and Alexander Semenov from MTS told about researching social network data.



After that, the guys united in teams and met with the curators, who were the young developers and graduates of ShAD Yandex. The teams could use any open data to implement their ideas, while the organizers provided datasets from Stepik, QIWI, HH, VK, Skyeng, MTS and GitHub.

Thanks to the generous partner of Servers.ru, each team had the opportunity to grab their own GPU and enjoy life.

The second day began with a master class on Deep Learning from Alexander Panin, working with texts from Alexander Petrov and with dirty data from Kirill Lunev from Yandex. There was also a presentation of a project about monitoring audience attention by video from students of the National Research University Higher School of Economics under the guidance of YDF.

Presentations of the lectures can be here , workshops - here .

Then all the participants plunged into the continuous convulsions of the birth of projects, periodically suffering from a lack of a specific task and a leaderboard, as well as enjoying the fun features of the site and the environment.


As it usually happens on hackathons, in the end everything ended with the project presentations from the teams that survived to this solemn moment.

It is worth noting that this time the final decision was made not only on the basis of a speech, as is usually the case, the score for the technical implementation from the curators was also taken into account. This added to the entourage of the process, the guys frantically made demos, notebooks and raked their govnokod.

As a result, the first place was shared by 2 teams that submitted the following projects:

- A team of star schoolchildren from Votkinsk, Moscow and Arkhangelsk with a chat bot ZuckWorkBot for the selection of online courses on Coursera meeting the requirements of the selected Facebook job. The guys pleased the jury with a fun presentation that everyone wants to work at FB, and told how they can help us all.

You can test the bot, look for @ZuckWorkBot in Telegram.

- A lonely first-year student with the FKN HSE with the Q & A matching project, the main idea of ​​which is to select an answer to a question posed by natural language. The entire system was trained in dataset questions from TopQuestion. Questions and answers are broken down into words and vectorized using Word2Vec, they fall into a recurrent neural network, and then with the help of triplet loss they are displayed in a high-dimensional vector space. When a new question is asked, the same transformation is applied to it, after which the L2-nearest answer is searched. The results can be used to automate the support of software products, online courses and so on.

The second place was taken by the guys with the project Personnel on the assessment of wages on the job description and skills required from the candidate. The Skill2Vec system is implemented by converting skills into vectors like Word2Vec and processing natural language with recurrent neural networks. One of the funniest examples: “programmer” - “analytical thinking” = “1C programmer”, a demo here .

Third place is the team that introduced the StudentClass system to track student progress on the Skyeng English language learning online platform during voice communication with the teacher. With the help of neural networks, the audio recording of the lesson determines who is currently speaking and in what language.

Special prizes in nominations from QIWI were also awarded. The company not only provided its data and tasks for them, but also rewarded 3 teams according to the results: AlexNet, Untitled1 and Amise.

Comments on tasks from QIWI:

“Last year there were studies that showed that students do not find work after graduation. We looked at a lot of students playing games. And where the games - there and QIWI. We looked at all the data on payments, presented how they can be analyzed. The task was to investigate the data and find in the payment history some set of signs indicating that the student had graduated from the university and had a profit. For example, began to pay for utilities or spend more. The participants liked the task, because it is very similar to the real set of tasks that research companies cooperating with the largest European banks solve. They are also looking for events on the payment history - they discover a wedding, a quick birth of a child, a car purchase, - and the bank offers the client appropriate loans and other products.
The second task was to determine the socio-demographic characteristics of the user. The team that won the first prize, with high accuracy, learned how to determine gender and age, and their solution can be truly applied in business.

Of course, the guys expected that the banks a lot of data. But in fact, the payment history of the bank looks pretty boring: the category of the transaction, its amount and time. But from this boring list, you can select a huge number of signs. And we wanted the guys to try this work on their own.

We would also like to note that on the hackathon the competition itself often faded into the background.
Although the teams were rivals, they helped each other a lot, shared pieces of code, tested them. On adult hackathons this does not happen at all. Generation Z — born after 1995 — generally has increased team building skills and role assignments. And more ability to achieve a common result. "



Also, the MTS and Rybakov Foundation established its nominations with cash grants, Innopolis University with grants for free education in the bachelor’s program, FNS HSE with an invitation to the Bayesian School of in-depth training. And the company Profilum, for example, invited the winning teams for dinner with the investor Gor Nakhapetyan.

According to the results, the organizers chose 6 schoolchildren, who will go to the final of the NTI Olympiad, bypassing the qualifying rounds. By the way, at the GoTo winter school there will be a separate mini-group to prepare for the Olympiad in the direction of Big Data and Machine Learning.

The Russian venture company has become the general partner of the hackathon, the Servers.ru company has become the technical partner.

Comments of Alexander Petrov, one of the jury members:

“I was pleasantly surprised by the level of projects on the hackathon - it corresponded to the most“ adult ”level. The guys used the methods and technologies that have appeared recently and are just beginning to find application in real business, for example, the technology of training deep neural networks. I am sure that the knowledge and experience gained in the framework of the hackathon will be useful to everyone for future projects. We from E-contenta will definitely continue to cooperate with GoTo, since for us it is not just charity: already two guys from GoTo have completed an internship with us, and their code works in combat mode on large projects. ”


Learn more, as well as apply for participation in the next hackathon here .