Talks

The speaker will tell how his team wrote a wrapper for pytest and now run different types of tests for different products: from different teams on different hardware in SDL requirements.

The talk will cover the architecture of the tool, the implementation of some of its parts, and will show how it works for different user scenarios.

Have you had each model developed by your team in its own unique repository, unlike the others? And was it that each of the models had its own versioning and output system, a different way to run training, and even a different way to manage hyperparameters? On the speaker's team, there was. And it hurts.

From the talk you will learn how the team managed to unify the development of all the models and make a convenient framework for it. The speaker will tell you what difficulties they encountered, how they were solved, what the result was and what useful things they learned in the process.

We take stock, remember the bright moments and talk about our plans. Join the broadcast, so you don't miss anything!

Using a live project as an example, the speaker will deploy and configure a Django application from scratch for easy development. He'll make it 12-factor, bolt on linters, use actual toiling for testing, and put it all in CI.

When preparing data for machine learning, it is often necessary to perform a series of transformations on the data to prepare a dataset for training.

We will talk about a small library developed by the Microsoft Commercial Software Engineering group, which allows you to describe data processing as a single pipeline with named data streams. With the help of such a library it's convenient to process data that are too big to fit into Pandas DataFrame, but too small to use Spark/Databricks.

Speakers will tell why in their team decided to develop their own open source framework for working with time series in Python, what pains they experienced during the development process and how they tried to solve them.

The problems they will talk about are not tied to the specifics of their framework and can occur in any open source project. In this sense, the talk will be interesting for a wide range of developers and will be both instructive and entertaining.

Let's discuss with the speakers (and not only) what modern companies are replacing Python in their work.

In the PiterPy 2022 Main Studio, we will analyze the questions from the quiz from CROC.

We will talk about the schedule, sessions, and share the information. Join the broadcast to find out what's on the air soon!

In 2016, Anton started writing a single-player browser game based on Travian and Heroes in JS. Five years and 30K SLOC later, he decided to add common events for all players. And so it went: the announcement system, the table of records, the events themselves, the distribution of rewards. Everything was done in a simple way, on Flask. Then the chat provider "tightened the screws" and before the release of the beta had to write his own chat. 
On the day of release, 31.12.2021, Anton found out how bad the idea was :)

We will talk about the schedule, sessions, and share the information. Join the broadcast to find out what's on the air soon!

Using a live project as an example, the speaker will deploy and configure a Django application from scratch for easy development. He'll make it 12-factor, bolt on linters, use actual toiling for testing, and put it all in CI.

Live from the main studio of PiterPy 2022.

Live from the main studio of PiterPy 2022.

Vladimir will talk about parametric generation of autotests using Pytest. As part of the talk during the livecoding session, he will show how to write the simplest parametric tests. The speaker will also demonstrate a few tricks to make tests more readable and understandable to a third party, as well as mark some of them grayed out when dropped (without blocking the build).

The speaker will talk about his experience of porting GAN model to run on a mobile device. The problems to be faced, as well as options for solutions, will be discussed.

In this talk, the speaker will answer the questions:

• what are the main components of a 3D scene and what is rendering;
• what internal and external camera parameters are responsible for and how they affect the final image;
• How to render using Pytorch3D;
• how to determine the camera position in 3D space from a single photo and 3D object model using Pytorch3D;
• features of implementing the same solution using TensorFlow Graphics;
  what tricks can be used for better convergence of the algorithm.

Credit scoring is a classic task for a bank. Over the years, tons of tools have accumulated to solve it. How do you choose the best one? Most methods rely on classic machine learning models, so you are guaranteed to lose information when working with sequential data.

The speaker will tell you how to bypass these risks and improve product metrics in a credit scoring task using neural network approach.

William will showcase and go through different steps to analyze Stack Overflow questions about certain topics using an automated pipeline with CI/CD principles.

Python is widely used for mathematical calculations. In many cases, performance is a critical issue in such calculations. Any modern computer is equipped with a graphics processor. Often it has significantly better computational performance than the CPU. You'll learn how and in what situations you can quickly incorporate the power of the GPU into your Python code in your calculations without having to program the graphics cards yourself.

Have you ever wondered how to run multiple isolated Python interpreters with limited resources? When you look deeper into this question, it is very non-trivial.

When the standard methods fail to work, you may want to consider a tool from the world of web applications - the WebAssembly virtual machine. In his talk, Alexander will tell you how he runs Python on Wasm and explain for which tasks this will be most useful.

We've all read Robert Martin's book on clean code and love PEP8. And it's true: many people follow best practices in their development. But are they all the best, and who are they all?

The speakers have analyzed over five thousand popular open source projects and are happy to share the results with the community. Let's talk about them both in terms of analysis methodology and interesting findings.

We will talk about the schedule, sessions, and share the information. Come to the room or join the broadcast to find out what to expect soon!

The researchers at European University (and not only there) face large inconveniences when working with microfilms, reduced copies of paper documents on film. They addressed the problem to the director of the Center for Machine Learning. There was developed an algorithm which allows you to convert the video of browsing the slides on microfilm into a ready PDF. Telegram bot was also designed for users.

Technologies: ML and neural networks (PyTorch, YOLO, DBSCAN), libraries for working with computer vision (OpenCV, sciPy), AIOGram bot. 

Target audience: people who want to do computer vision, but do not know what to begin with, experts in CV and its researchers.

An online programming course, a university course, a job interview... Everywhere you need fresh code tasks. The ideal is to equip everyone with a personalized set of tasks. But what to do if there are too many students? There are some systems which provide automatic checking of tasks, but what about systems which can automatically compose interesting programming tasks in necessary quantity? The development of such a system is what we will talk about.

Live from the main studio of PiterPy 2022.

Кто вообще придумал эти микросервисы? Python и так медленный, а теперь будет еще и больше потребления ресурсов, потому что для каждого микросервиса нужна память и хост.

В докладе спикер покажет на типичных кейсах, как можно сделать монолитное приложение и как решить те проблемы, которые мы хотели избежать, переходя на микросервисы.

The speaker will tell you how to find highlights in Dota 2 matches. To do this, we will get acquainted with the structure of replays and implement an algorithm for finding interesting moments.

The talk will be held in an interactive format, the code will be posted in the public domain. For preliminary acquaintance you can read the series of articles on Habr.com.

We take stock, remember the bright moments and talk about our plans. Come to the room or join the broadcast, so you don't miss anything!

Just as the work on AlexNet in 2012 made a significant breakthrough in the field of CV, Annention Is All You Need in 2017 allowed us to assess just how deep the rabbit hole is for the world of NLP.

Let's discuss the past, present and future of this ML trend with natural language processing experts.

Is it time to give up on classic feature extraction from texts in the age of Transformers? How suitable are NLP methods for formal programming languages? How to take the Turing test now without registration and SMS and much more.

There are dozens of services with ML models in production at Avito, and Oleg's team has long experience of saving hardware for them. Oleg has some insights about how to save up to 200% servers, and he is ready to share them (insights, not servers).

Live from the main studio of PiterPy 2022.

Let's look at Python as a functional programming language and also at its potential pitfalls and advantages. There will be examples both using standard libraries and the most interesting third-party ones.

The talk will cover practices and tools that will allow you to study system behavior in a productive environment, making it easier and faster to diagnose and solve problems. Examples will be shown in Django and Celery-based system, but with minimal modifications they can be used with other frameworks.

The text of calls is an important data resource on which you can build many models and raise a huge number of hypotheses. The only "but" is that it is important to post-correct speech recognition errors and not to forget about the two channels.

Alexey will discuss interesting vulnerabilities and bugs that are found in the architecture of Python applications. The talk will cover business logic vulnerabilities, type customization, races, templates, libraries, and more.

The speaker will tell how he and his team made the Proof of Concept of a streaming ELT pipeline for a large American insurance company.

At the master class you will get acquainted with reinforcement learning, learn the possibilities of application, advantages and disadvantages of this direction. You will get acquainted with the basic RL algorithms and build your first model.

Decentralized exchange Uniswap is one of the brightest representatives of Web3 and DeFi. From the talk you will learn how a decentralized exchange is organized and how you can work with it using Python.

Alexey will talk about:

• his experience in implementation of different MVP projects;
• what's the difficulty of their launch at microservices;
• what problems will be faced by the developers at start;
• which solutions need to be considered immediately and which ones may be postponed;
• what needs to be worked out and considered when planning deadlines.

Live from the main studio of PiterPy 2022.