Machine Learning Boot Camp IV. Fourth. Secret. Your
On April 21, we open the fourth machine learning competition on the ML Boot Camp platform. Today we will talk about the new task, updates on the site and other useful nishtyak. And if you suddenly hear for the first time what ML Boot Camp is, go under the spoiler and we'll tell you everything.
ML Boot Camp is a platform for solving machine learning problems. Periodically, we post new tasks on it and launch a contest. Participants must solve our problem within a month and send a solution. The authors of the best decisions will receive prizes. In the last championship, we gave MacBook Air for first place, iPad for second and third and iPod nano for 4-6 places.
At the start, the participants receive the conditions of the problem, a verbal description of the available data - a training sample. The sample consists of marked examples - vectors of descriptions of each object with a known answer. Participants use computer-assisted learning methods known to them. They use the trained system on new objects (test sample), trying to determine the answer for them.
The test sample is randomly divided into two parts: rating and final. The overall result on the rating data is calculated by the system and published immediately, but the winner is the one who gets the best results on the final data. Results remain hidden for participants until the very end of the competition.
On the last day of the championship the participant can choose two decisions that will represent him in the final. The best of them will count on the leaderboard.
New task
This time we offer you a "Task with a secret". We will not reveal the substantive formulation of the problem. She will remain unknown until the end of the competition. You will be able to test your analytical skills in full!
You are faced with the task of classification: based on a well-known distribution of five classes of teaching elements, distribute test ones. As a response, send a text file, each line of which corresponds to a line in the file with test data and contains the class number (0, 1, 2, 3 or 4). We offer you as many as 42 numeric characters for classification!
The criterion for the quality of the decision will be the proportion of correctly classified objects. The test sample is randomly divided into two parts in a 40/60 ratio. The result of the first 40% will determine the position of participants in the rating table throughout the competition. The result for the remaining 60% will be known after the end of the competition and will determine the final placement of participants. Good luck!
Many thanks to UNN them. NI Lobachevsky and personally Nikolai Zolotykh and Oleg Durandin for their help in preparing the task and expert support of the championship! Nikolay and Oleg participated in the conduct of each of our ML championship, without them we would not have mastered half of what has been done now.
Useful materials
Tutorial article
If you are a beginner, we recommend that you read a small tutorial on our platform. In it, you will analyze the “Credit Scoring” task and learn how to predict whether a loan will return to the bank according to the client.
The article contains extracts of test data, their visualization, pieces of Python code and all semantic conclusions.
Parse ML BootCamp I
On ML Boot Camp we already worked with anonymous data. In a closed student contest, we were asked to classify binary sequences. The quality criterion was also the proportion of correct answers. Pavel Shvechikov achieved an impressive 0.6785, provided that some sequences were written by people, the second - a random number generator, and the third - an algorithm.
We asked the contest winners to tell the main ideas of their decision and collected them in a separate publication on Habré . Perhaps their ideas will help you choose the direction of travel. Look, there are cool visualizations there:
You can visualize everything. Even binary sequences
Sandbox
You can practice before the start of the championship, including the “Binary trees” problem in the Sandbox . There are available any tasks of past championships, you can download your solution and find out the rating. For each task in the sandbox has its own leaderboard. If a new task seems to be too difficult (or, on the contrary, simple) - conquer the rest.
In the sandbox you can solve all the problems of old contests.
Chat in Telegram
Now, thanks to the official championship chat, you can ask your question directly to the organizers. You can also ask for advice or share insights on the decision. All participants gather here and storm the task. They will help you with fresh ideas and kind words.
Experienced controllers participate in the chat, including winners of past contests.
Forum
The participants very much asked for a forum for putting there something that could easily get lost in the chat. It’s not a fact that we will have time to open it just for the start of the competition, but we can definitely promise that in the near future the forum will appear on our platform.
Two solutions as answer
From this point on, you can choose two solutions as the final answer. What scores a larger Score in the final sample will be your result in the championship. This will help you, for example, if in one of the solutions you have a more stable model, but the other gives the best result on the test sample.
Prizes
This time we will break the slender ranks of Apple technology in the prize pool. For the first place we will present the MacBook Air 13 laptop, for the second and third - the Samsung Gear S3 Frontier smart watch. If you don’t enter the top three, but enter Top 6, you’ll be friends with a 6TB WD My Cloud personal cloud drive. And, as always, the TOP 50 championship participants will get T-shirts with the championship logo.
check in
The championship will open on April 21 at 14:00 Moscow time. You can register on the platform at this link . Until the movement has begun, come to solve problems in the Sandbox .
')
Source: https://habr.com/ru/post/326860/
All Articles