Multi-armed bandits in the recommendations

Hello! My name is Misha Kamenshchikov, I am engaged in Data Science and the development of microservices in Avito's team of recommendations. In this article, I'll talk about our recommendations for similar ads and how we improve them with the help of multi-armed bandits. With a report on this topic, I spoke at the conference Highload ++ Siberia and at the event «Data & Science: Marketing» .
 
Multi-armed bandits in the recommendations Recommendations for Avito
 
Service similar ads
 
Rankers and configs
 
A /B tests ...
+ 0 -

angular-ngrx-data - state management and CRUD for five minutes

angular-ngrx-data - state management and CRUD for five minutes
 
To date, no single large SPA the application does not do without state management . For Angular in this direction there are several solutions. The most popular of them is NgRx . It implements Redux Pattern using the library RxJs and has good tools.
 
In this article, we briefly go through the main modules NgRx and in more detail we will concentrate on the library angular-ngrx-data , which allows you to make a full ...
+ 0 -

Deep Learning: Recognition of scenes and attractions on images

Deep Learning: Recognition of scenes and attractions on imagesTime to replenish the treasury of good Russian-language reports on Machine Learning! Moneybox itself is not replenished!
 
 
This time we will get acquainted with the fascinating story Andrei Boyarov about the recognition of scenes. Andrey is a programmer-researcher engaged in computer vision at Mail.Ru Group.
 
 
Scene recognition is one of the most actively used areas of computer vision. This task is more complicated than the studied recognition of objects: the scene is a more complex and less formalized concept, it is more difficult to identify the signs. From the recognition of scenes ...
+ 0 -

We are looking for speakers at Moscow Data Science Major

We are looking for speakers at Moscow Data Science Major
 
 
September 1 Mail.Ru Group and Open Data Science community will hold the largest Moscow Data Science Mitap.
 
 
We will open a new academic and working year for the whole day of sections and networking!
 
Fill in the form , to perform at the Mitap. Please note that completing the form does not guarantee your participation. We will review all applications and reply to the mail you indicated about the decision.
 
 
Mitap is focused on Data Science professionals and enthusiasts. We do not accept marketing or advertising reports. It is important for us that the reports are relevant and useful to the ...
+ 0 -

Automation of salary monitoring using R

Automation of salary monitoring using RDo you know how much you are underpaid? And maybe overpay? What is the ratio of resumes and vacancies to a position that is similar to yours?
 
Answering this question, you can lie to yourself, you can brazenly lie, but you can operate with statistics.
 
In fact, every self-respecting office regularly conducts monitoring of salaries in order to orientate in the labor market segment of interest to it. The results of such monitoring help in the formation of the wage policy, in the drafting of the budget of the Labor Code and in the adoption of a whole range of these strategic decisions.
 
Despite ...
+ 0 -

Truth and lie of face recognition systems

Perhaps there is no other technology today, around which there would be so many myths, lies and incompetence. The journalists who tell about the technology are lying, politicians who talk about successful implementation lie, most tech vendors lie. Every month, I see the consequences of how people try to introduce face recognition into systems that can not work with it.
 
Truth and lie of face recognition systems  
The topic of this article long ago piqued, but it was all somehow lazy to write it. There is a lot of text that I've repeated to different people twenty times already. But, after reading the next pack of trash still decided it was time...
+ 0 -

Call of Dates, or How the second hackathon Mail.Ru

July 14-15 Mail.Ru Mail held its second hackaton SmartMail Hack 2: Call of Data. Participants needed to develop a new, interesting and useful solution for submitting data from the Post. And we are in a hurry to tell you about the winners of projects on Hakaton.
 
 
Call of Dates, or How the second hackathon Mail.Ru  
e.mail.ru/settings/cleanmaster . The guys offered to create an application for easy viewing, sorting and unsubscribing from newsletters. And the interface worked very well, for which they received a maximum score of 3 points:
 
 
 
 

Conclusion


 
Certainly, i...[/h]
+ 0 -

How we won in SmartMailHack 2

How we won in SmartMailHack 2Last weekend (July 14-15) in mail.ru the next hackaton SmartMailHack was held. We were asked "to come up with a feature that will allow access to data from the Mail and more efficiently interact with them."
 

Description of the data, ideas and solutions


 
We received a test box with more than 1500 messages, as well as full access to it through the API. The organizers provided a large, detailed manual on its use (a book of 547 pages). Using a token and simple JSON requests, we could receive all the necessary information about the mail: letters, sender names, various characteristics.
 
After ...[/h]
+ 0 -

To within a hundredth: top-10 reports SmartData 2017

To within a hundredth: top-10 reports SmartData 2017  
 
Viewers of the SmartData conference are people who like to work with data. It must be assumed that they assessed the evaluation of the reports after last year's conference very thoughtfully.
 
 
And now, according to these estimates, we made the top 10 video recordings. And at the same time, to please the fans of data, they indicated all the accompanying figures for each of the ten reports: the place in the top, the exact spectator rating, the number of spectators.
 
 
Generally speaking, often the neighboring positions in the top ratings are slightly different. So, perhaps, it is not worth attaching ...
+ 0 -

Neurobugurt. How we taught the neural network to come up memes a year before Stanford

Neurobugurt. How we taught the neural network to come up memes a year before StanfordBy writing this article I was pushed here by this news (+ 3r3r3? study
) About the invention of a meme generator by scientists from Stanford University. In my article, I will try to show that you do not need to be a scientist from Stanford to do interesting things with neural networks. In the article, I describe how in 2017 we trained a neural network on the body of about 3?000 texts and forced it to generate new Internet memes and memes (communication signs) in the sociological sense of the word. The algorithm of machine learning, technical and administrative difficulties that we have encountered...
+ 0 -