On October 9, students, postgraduates and young teachers of the SAKMIG department joined the Kaggle World Competitions in Artificial Intelligence. The formation of 3 teams with the following preliminary names has begun: - SAIT-1 VNTU (graduate students of specialties 124 and 126 and 1st year undergraduate students from 126 (specialization "Applied IT"), captain - 3rd year postgraduate student from 124 Mykhailo Dratovany,
– SAIT-2 VNTU (students of all our 1-4 years of undergraduate studies 124, 2nd year of 126 (specialization "Applied IT") and master's degrees 1-2 years of 126, captain - associate professor, Ph.D. of 126 Andriy Yascholt),
SAIT-3 VNTU (4 3rd year students and 1st year students out of 124, captain – 3rd year student out of 124 Ivan Klishyn).
Captains Mykhailo Dratovany and Ivan Klishyn have already participated together with Prof. Mokin V.B. in the previous Kaggle competition on predicting the cost of taxi trips. In addition, all teams have a quota of up to 2 places for 11th grade students who will study at the SAKMIG department courses on applied IT artificial intelligence. Other participants must be only from among students of different levels of education of the SAKMIG department. They will receive assistance from candidates of sciences in specialty 126 from associate professors of the SAKMIG department: Andriy Yascholt, Ilona Varchuk, Oleksiy Kozachko.
The IT company “NestLogic” has proposed to start the competition to predict the cost of purchases in Google online stores around the world “Google Analytics Customer Revenue Prediction”, guaranteeing prizes to the best teams from Vinnytsia. Kaggle provides all participants with a csv file with about 1 million records (with purchases from about a thousand cities, including the cities of Kyiv, Lviv, Kharkiv, etc.) with more than fifty parameters in each. You need to learn to predict the cost of purchases by each user, and this control sample is more than 800 thousand records. There are a number of additional conditions. The task is quite complex and non-trivial.
Given the lack of experience of the participants in the Python programming competition, Prof. Mokin V.B. helped them write a draft of the program code that reads all parameters, including from JSON format, optimizes their data types and the list of parameters (this allowed to reduce the volume of training and control data files by 7 times from 1.4 GB to 200 MB, which significantly speeds up processing), identifies a typical high-speed LG-boosting model with 5-fold cross-validation, combines all store visit sessions for each user and calculates a logarithmic cost estimate, in accordance with the requirements of the task. No processing and clustering or classification of the input data was carried out - this should be done by the participants of the competition. To check the operability of Prof. Mokin V.B. uploaded the prediction result to Kaggle – it was expected to be among the last of 2309 teams, although, surprisingly, it overtook more than 300 teams and is only 0.3% away from 3rd place.
Meanwhile, Prof. Mokin V.B. plans to write a normal program with variable clustering and other techniques in parallel with the students. After this competition ends on November 15, a joint seminar on the exchange of experience will be held. Since the main thing in this is to learn how to correctly apply artificial intelligence technologies in practice. Then there will be another competition.
Attention, students, the formation of teams continues until November 7, but you can join the team only by giving it some idea that will allow it to rise in the ranking - one desire is not enough. Students who want to join can directly contact the captains or Prof. Mokin V.B. via Facebook messenger. After November 15, there will be a new set.