Report from Moscow Data Science Meetup May 31

On May 31, the Moscow Data Science Meetup gathered more than 200 participants in our office. At the meeting, we talked about the gradient boosting, baseline on ConvAI.io and dismantled the case, which received the 7th place out of 419 teams at the competition Dstl Satellite Imagery Feature Detection. We bring to your attention video recordings and presentations of three reports presented at the meeting.

Alexey Natekin (DM Labs, OpenDataScience): “Gradient Boosting: Features, Features, and Chips Outside of Standard Kaggle-Style Tasks”

Many people use gradient boosting, but few people read documentation on it. Even fewer people are interested in articles and recipes, how to prepare it better and what to do with it. During this report we will just go through a set of interesting features, tricks and recipes.

Valentin Malykh (Laboratory of Neural Systems and Deep Learning of MIPT): “How to stop being afraid and start to solve convai.io”

For those who wanted to participate in the competition to create conversational artificial intelligence, and for those who wanted, but did not know about it, it will be interesting to look at how the baseline on convai.io works. We will also give an overview of the current state of the area, to understand where we are, and why this competition should be held right now.

Eugene Nekrasov (Mail.Ru Group): “Solution of the Dstl Satellite Imagery Feature Detection (Kaggle) task”

In the Dstl Satellite Imagery Feature Detection competition, participants were given the task of segmentation of multispectral satellite images. The report will analyze the solution to this problem based on convolutional neural networks, which was awarded 7th place out of 419 teams.