Welcome to Sberbank Data Science Journey 2018 - a race for machine learning algorithms
This fall, we will once again organize a large machine learning competition, Sberbank Data Science Journey. Every year we cover some new topic and now we invite you to try our hand at AutoML. More specifically, in the development of a very skillful meta-algorithm that is able to independently create machine learning models: with data processing, feature construction, model training, selection of their parameters and prediction of the target variable.
This year, we will award 13 teams for solving the problem. The rest of the details - further in the post.
The conditions for this year’s task are set out in detail on GitHub . In short, you need to create a machine learning algorithm that automatically performs data preprocessing, selection of a family of models, as well as selection of hyperparameters. And, of course, meets all the conditions. To work, in addition to these conditions, given a public set of dataset .
')
Each team of participants will be provided with a personal account on the competition website. A team may include a maximum of four people. During the competition, from September 19 to November 3, participants will be able to upload up to five algorithm solutions daily to their personal area. All these solutions fall into the overall rating, where they are regularly checked online for a set of closed test data. From the results, a table of overall results is compiled, open to all - you can always estimate your success. Download solutions will be available until November 3, 11:59:59 PM (hereinafter, Moscow time).
Among all their loaded decisions, each team will have to choose two, which will present it in the final. This is additionally given 12 hours. Deadline of choice - November 4, 12:00 .
And finally, the next 12 hours we will drive the data on the final decisions and announce the winners on the competition website by the end of this period.
The developers of the ten decisions with the highest final rating will be awarded cash prizes: 1,000,000 rubles for the first place, 500,000 for the second, 300,000 for the third, 200,000 for the fourth and fifth. All other places up to the tenth inclusive will receive 100,000 rubles each.
In addition, if the team is not too lazy to publish the decision on GitHub, you can get another 100,000 rubles. We will play three such awards among the best solutions that will be laid out for public use. And by the way, this additional prize does not exclude the main one.
The official award will take place on November 10 at the Sberbank Data Science Day conference. On the confe we will write further. We will undertake the transportation costs of developers from other cities, but even if this does not lure you to the capital, we will still pay the prize.
In case of different misunderstandings, all the information about the event is available in a dry legal language in a separate document . If you have not found the answer there or the question concerns the technical part, post it on the forum or in the comments to the post. We will be glad to your participation!
Useful links:
This year, we will award 13 teams for solving the problem. The rest of the details - further in the post.
Terms and conditions
The conditions for this year’s task are set out in detail on GitHub . In short, you need to create a machine learning algorithm that automatically performs data preprocessing, selection of a family of models, as well as selection of hyperparameters. And, of course, meets all the conditions. To work, in addition to these conditions, given a public set of dataset .
')
Each team of participants will be provided with a personal account on the competition website. A team may include a maximum of four people. During the competition, from September 19 to November 3, participants will be able to upload up to five algorithm solutions daily to their personal area. All these solutions fall into the overall rating, where they are regularly checked online for a set of closed test data. From the results, a table of overall results is compiled, open to all - you can always estimate your success. Download solutions will be available until November 3, 11:59:59 PM (hereinafter, Moscow time).
Among all their loaded decisions, each team will have to choose two, which will present it in the final. This is additionally given 12 hours. Deadline of choice - November 4, 12:00 .
And finally, the next 12 hours we will drive the data on the final decisions and announce the winners on the competition website by the end of this period.
Awards
The developers of the ten decisions with the highest final rating will be awarded cash prizes: 1,000,000 rubles for the first place, 500,000 for the second, 300,000 for the third, 200,000 for the fourth and fifth. All other places up to the tenth inclusive will receive 100,000 rubles each.
In addition, if the team is not too lazy to publish the decision on GitHub, you can get another 100,000 rubles. We will play three such awards among the best solutions that will be laid out for public use. And by the way, this additional prize does not exclude the main one.
The official award will take place on November 10 at the Sberbank Data Science Day conference. On the confe we will write further. We will undertake the transportation costs of developers from other cities, but even if this does not lure you to the capital, we will still pay the prize.
In case of different misunderstandings, all the information about the event is available in a dry legal language in a separate document . If you have not found the answer there or the question concerns the technical part, post it on the forum or in the comments to the post. We will be glad to your participation!
Useful links:
Source: https://habr.com/ru/post/424635/
All Articles