PyData
76 talks from the 2019 edition.
-
A Bayesian Workflow with PyMC and ArviZ ▶ Recording 📝 Transcript
Corrie Bartelheimer
Bayesian Modelling has several advantages such as the handling of uncertainty. While the advantages are well known, implementing a Bayesian model can be a bit more involved and some care needs to be taken to check...
-
A Tour of JupyterLab Extensions ▶ Recording 📝 Transcript
Jeremy TULOUP, QuantStack
JupyterLab can be extended via third-party extensions written by developers from the Jupyter community. This is a tour of 20 of these extensions, in 20 minutes. Demos included!
-
Active Learning with Bayesian Nonnegative Matrix Factorization for Recommender Systems
Gönül Aycı
In most of the systems, collecting data is not always free. I will talk about an approach for a matrix completion problem that learns a distribution of data where information is incomplete or collecting it has a cost.
-
AI Intentions and Code Completion ▶ Recording 📝 Transcript
Vasily Korf
Datalore supports intentions – code suggestions based on what you’ve just written. They cover a wide range of situations from generating code to warnings and optimization suggestions.
-
Airflow: your ally for automating machine learning and data pipelines
Enrica Pasqua, Bahadir Uyarer
Now that you finally have your Machine Learning model trained, what’s the next step for moving to production? Orchestrating, scheduling and monitoring ML inference pipelines is a big challenge. Airflow can be your...
-
Algo.Rules - How do we get the ethics into the code? Keynote ▶ Recording 📝 Transcript
Carla Hustedt
In the keynote I will present our Algo.Rules, 9 rules for the design of algorithmic systems and address the questions “What standards of quality should algorithms be held to?” and “How can we make sure that these...
-
Applying deployment oriented mindset for building Machine Learning models
Marianna Diachuk
Developing ensemble model with hundreds of features? And getting stuck for months trying to deploy the model and fighting with data inconsistency and bugs? This talk will introduce the way to build the development...
-
Are you sure about that?! Uncertainty Quantification in AI ▶ Recording 📝 Transcript
Florian Wilhelm
There is a strong need in many AI applications to state the certainty about their predictions. This talk elaborates on different ways to perform uncertainty quantification in deep learning and classical methods.
-
Automated Feature Engineering and Selection in Python ▶ Recording 📝 Transcript
Franziska Horn
Careful feature engineering and selection can be just as important as choosing the right ML model & hyperparameters. I will present several options for automating the feature engineering and selection process with a...
-
Automating feature engineering for supervised learning? Methods, open-source tools and prospects.
Dr. Thorben Jensen
Feature engineering is crucial for supervised learning (SL), but labor-intensive. Automating this task is less established, compared to other steps of SL. We therefore present and compare approached and open-source...
-
Avoiding ML FOBO ▶ Recording 📝 Transcript
Rachel Berryman, Dânia Meira
Everyday there is a new package or algorithm to use- it can be hard to determine what is useful and what is only hype. The speakers offer a practical roadmap and checklist to help you cut through the hype and focus...
-
Birds of a feather flock together - Tracking pigeons with Python and OpenCV
Neslihan Edes
In this talk I want to demonstrate how to use OpenCV to implement basic animal movement tracking use cases. And everything without any fancy machine learning or neuronal networks ;)
-
Build a Machine Learning pipeline with Jupyter and Azure
Daniel Heinze
With increasing focus on Machine Learning systems in almost every business it is important, to build a great pipeline to train, test and deploy your models. In this session we will show a way to do that with Jupyter and Azure
-
CANCELED: Create CUDA kernels from Python using Numba and CuPy.
Emergency Self-Constuct
Get well, soon, Valentin! We'll explain how to do GPU-Accelerated numerical computing from Python using the Numba Python compiler in combination with the CuPy GPU array library.
-
CANCELLED: First steps in Julia
Felicia Burtscher
Cancelled - get well soon, Felicia! The community of Julia has been growing and the much anticipated 1.0 release of Julia is out since last summer. Targeted to Julia beginners and Python users, we will especially...
-
Creating an Interactive ML Conference Showcase ▶ Recording 📝 Transcript
Harald Bosch
Our goal is to create a simple yet interactive showcase for computer vision using a Python notebook. In a trade fair setup, we want to learn new object classes quickly using very few training examples. Thus, we rely...
-
Current affairs, updates, and the roadmap of scikit-learn and scikit-learn-contrib
Adrin Jalali
As a scikit-learn core developer, I'd give an update on recent changes, current affairs, and the roadmap of the package and the community packages included in scikit-learn-contrib. I'd also briefly talk about how new...
-
Dash: Interactive Data Visualization Web Apps with no Javascript ▶ Recording 📝 Transcript
Dom Weldon
Interactive web pages and visualizations with no JavaScript? What could go wrong? What you can, can't, should and probably shouldn't do with plotly/Dash.
-
Data Literacy for Managers ▶ Recording 📝 Transcript
Alexander CS Hendorf
Artificial Intelligence need to be better understood in enterprises. Close the communications gap between engineers and management. Making data litteracy happen in your organisation.
-
Decentralized and Privacy-Preserving ML via TensorFlow Federated ▶ Recording 📝 Transcript
Peter Kairouz, Amlan Chakraborty
Federated Learning is a technology to train machine learning models and run data analytics on decentralized data. This tutorial will demonstrate step-by-step how to train largescale TensorFlow models and custom...
-
Deep Learning for Healthcare with PyTorch
Valerio Maggio
This tutorial provides a general introduction to Deep Learning using PyTorch with specific focus on challenges and solutions for Healthcare and Computational Biology.
-
Detecting and Analyzing Solar Panels in Switzerland using Aerial Imagery
Martin Christen
We present a novel method for detecting solar panels and its geometry on aerial imagery. The goal is to know the exact locations, dimensions and potential of every solar installation in Switzerland.
-
Does hate sound the same in all languages? ▶ Recording 📝 Transcript
Andrada Pumnea
How might we make social media safer and more inclusive? Tackling hate speech online is not easy, especially if it’s in a language less circulated. This talk describes detecting hate speech in Romanian from dataset...
-
Embrace uncertainty! Why to go beyond point estimators for valuable ML applications
Stefan Maier
Usually, uncertainties of Machine Learning predictions are just regarded as a sign of poor prediction accuracy or as a consequence of lacking input features. This talk illustrates how modeling uncertainties can...
-
Equivariance in CNNs: how generalising the weight-sharing property increases data-efficiency
Marysia Winkels
In this talk, we will explore how the weight-sharing property of the convolutional layer can be generalised to achieve equivariance towards transformations beyond just translation, how to implement this, and the...
-
Extended Ligthning Talks CANCELLED: Crunching Numbers Like a Journalist Keynote
Marie-Louise Timcke
Get well soon Marie-Louise! As our world becomes more and more data-driven, journalists are equipping themselves with data science and programming skills to turn numbers into headlines.
-
Fairness in decision-making with AI: a practical guide & hands-on tutorial using Aequitas ▶ Recording 📝 Transcript
Pedro Saleiro
In this tutorial, we are going to deep dive into algorithmic fairness, from metrics and definitions to practical case studies, including bias audits using Aequitas (http://github.com/dssg/aequitas) in real policy...
-
Fighting fraud: finding duplicates at scale ▶ Recording 📝 Transcript
Alexey Grigorev
We present a duplicate detection system that we use to fight fraud in online classifieds. The system uses machine learning to analyze both text and images of 10 million ads daily and stop fraudulent listings before...
-
Friend or Foe: Comparison of R & Python in Data Wrangling & Visualisation
Yuta Kanzawa
R and Python are partially overlapping but different as a whole: community and language. Still, comparing them in their common fields such as data wrangling and visualisation, useRs and Pythonistas will deepen mutual...
-
Gaussian Process for Time Series Analysis ▶ Recording 📝 Transcript
Dr. Juan Orduz
The aim of this talk is to introduce the notion of *Gaussian process* and describe how to use it to solve regressions problems and time series forecasting.
-
Gaussian Progress ▶ Recording 📝 Transcript
Vincent Warmerdam
This talk is an attempt at explaining the power of the Gaussian[tm] by stepping up the ladder from Naive Bayes to Mixtures to Neural Mixtures to Gaussian Processes.
-
Get to grips with pandas and scikit-learn ▶ Recording 📝 Transcript
Sandrine Pataut
This session will be an exposition of data wrangling with pandas and machine learning with scikit-learn for Python Programmers. This hands-on workshop will cover a classification project, from importing the data to...
-
Hidden Markov Models for Chord Recognition - Intuition and Applications ▶ Recording 📝 Transcript
Caio Miyashiro
This tutorial describes the intuition behind Hidden Markov Models, with less mathematical formulas and with an application on Music Analytics - Chord Recognition
-
Hide Code, Minimize Dependencies, Boost Performance - The PyTorch JIT
Tilman Krokotsch
PyTorch makes developing, training and debugging deep neural networks convenient. Learn how to export your trained model using its just-in-time (JIT) compiler to hide your network architecture, minimize code...
-
How strong is my opponent? Using Bayesian methods for skill assessment
Darina Goldin
Being able to correctly estimate a competitors' skill is a crucial question in sport forecasting and matchmaking. This talk will provide an overview of the three most common algorithms for this task: Elo, Glicko2 and...
-
How to choose better colors for your data visualizations ▶ Recording 📝 Transcript
Daniel Ringler
Everybody is doing colorful charts with Python libraries such as matplotlib and bokeh but most people never change the basic configuration. This talk will teach you the basics of color theory to help you choose the...
-
How to write tests that need a lot of data? ▶ Recording 📝 Transcript
Sander Kooijmans
In this talk Sander explains how you can write unit and integration tests that need a lot of data. As an example Sander shows how to test code of a warehouse management system (WMS). This is not about big-data, though.
-
Interpretable Machine Learning: How to make black box models explainable
Alexander Engelhardt
Complex machine learning models make better predictions, but at the cost of turning into an unexplainable black-box model. In this talk, we'll look into a framework that allows us to explain why a model makes a...
-
Julia for Python ▶ Recording 📝 Transcript
Simon Danisch
Julia is a new Language, that is fast, high level, dynamic and optimized for Data Science. But due to its young age, it might not be for everyone yet. Learn about Julia's strengths and how you can integrate it in...
-
Kartothek – Table management for cloud object stores powered by Apache Arrow and Dask
Florian Jetter
Efficient data storage is an integral part of successful data applications. Cloud object stores prove to be an efficient choice but come with downsides when storing structured, tabular data. There is a way out, though.
-
Law, ethics and machine learning – a curious ménage à trois
Dr. Benjamin Werthmann
The talk addresses how law and ethics can provide a framework for the “machines on the rise” that ensures fairness and societal acceptance while not interfering (too much) with the drive behind the advancement of a...
-
Lessons Learned as a Product Manager in Data Science ▶ Recording 📝 Transcript
Tereza Iofciu
The fun part about data science is that no two people really agree on the definition of the data scientist role. So how does the role of product manager in data science look like?
-
Leveraging the advantages of Bayesian Methods to build a data science product using PyMC3 ▶ Recording 📝 Transcript
Korbinian Kuusisto
Bayesian models offer greater theoretical advantages compared to non-probabilistic methods, and also allow for more flexible model design. But how can one leverage these theoretical advantages to build a successful...
-
Loss Function Theory 101 ▶ Recording 📝 Transcript
David Wölfle
This talk covers the theoretical background behind two common loss functions, mean squared error and cross entropy, including why they are used for machine learning at all, and what limitations you should keep in mind.
-
Machine learning with little data - from digital twin to predictive maintenance
Andreas Hantsch
This talk is about the coupling of a digital twin model and a machine learning predictive maintenance algorithm in order to be able to detect anomalies in the operation of a not well-known hardware system.
-
Making the complex simple in data viz ▶ Recording 📝 Transcript
Tania Vasilikioti
Creating graphics that convey the desired message, are easily interpretable, but also beautiful can be a daunting task. This talk will demonstrate how to use *The Grammar of Graphics* framework to conceptualize the...
-
Managing the end-to-end machine learning lifecycle with MLFlow ▶ Recording 📝 Transcript
Tobias Sterbak
Machine learning requires experimenting with datasets, data preparation steps, and algorithms. Deploy models to a production system and retrain it on new data. MLflow is an open source platform for managing the...
-
Panel: Turn any notebook into a deployable dashboard ▶ Recording 📝 Transcript
Philipp Rüdiger
Quickly turn your existing analyses built on the PyData stack into shareable, standalone apps and dashboards using the new open-source [Panel](http://panel.pyviz.org) library without learning the details of web development.
-
Play Stupid Games, Win Stupid Prizes ▶ Recording 📝 Transcript
James Powell
This is reserved for a James Powell in-promptu talk, stay tuned! This is reserved for a James Powell in-promptu talk, stay tuned!
-
Practical DevOps for the busy data scientist ▶ Recording 📝 Transcript
Dr. Tania Allard
How many times have you developed a model or a data application and tested it locally or in a staging environment just to find out that it breaks in production? This is a common issue faced by thousands of data...
-
Privacy-preserving Machine Learning for text processing ▶ Recording 📝 Transcript
Sarah Diot-Girard
Privacy is something we all care about, but when it is time to put our principles into application, it is not so trivial, especially when working with text. This talk aims at presenting a few options to handle...
-
Production-level data pipelines that make everyone happy using Kedro
Yetunde Dada
Learn how easy it is to apply software engineering principles to your data science and data engineering code. Expect an overview of Kedro, a library that implements best practices for data pipelines with an eye...
-
Professional Development and Career Progression for Data Scientists ▶ Recording 📝 Transcript
Noa Tamir
In this talk you will learn how to level up your skills, and develop your your career using on the job opportunities, as well as open source contributions
-
Python-Powered OSINT! Modernising Open Source Intelligence for Investigating Disinformation
Mx Chiin-Rui Tan, Dare Imam-Lawal
OSINT, the discipline of gathering intelligence from open sources, is critical for state & citizen interest but has lacked modernisation. We present a socio-technical maturity model using Python to update legacy...
-
Quantum computing with Python ▶ Recording 📝 Transcript
James Wootton
Frameworks for quantum computing are a new way to use Python for cutting-edge science, and to plan for future applications of this new technology. This session will serve as an introduction to quantum computing as a...
-
Rethinking Open Source in the Era of Cloud & Machine Learning Keynote ▶ Recording 📝 Transcript
Peter Wang
By some measures, Open Source is a wildly successful and crucial part of many areas of modern technology. However, the ’sustainability crisis’ and the age of cloud computing have threatened its core mechanisms. Peter...
-
Should I stay or should I go? Optimal exercise decisions using the Longstaff-Schwartz algorithm
Benedikt Rudolph
This talk presents a Python implementation of the Longstaff-Schwartz algorithm for financial exercise option valuation. The technical problems of optimal exercise decisions and exercise option valuation are...
-
skorch: A scikit-learn compatible neural network library that wraps pytorch
Benjamin Bossan
This talk is about the open source package [skorch](https://github.com/skorch-dev/skorch), a wrapper library that allows you to combine the best of sklearn and PyTorch. It covers when it makes sense to use skorch and...
-
Tackle the problems that really matter - leverage the power of data science in the service of humanity
Eva Schreyer, Lisa Zäuner
Data Science for Social Good Berlin (DSSG) brings together data scientist volunteers and nonprofit organisations that need support in tackling various data challenges. In this talk we will tell you how data science...
-
Take control of your hearing: Accessible methods to build a smart noise filter
Peggy Sylopp, Aislyn Rose
Have you ever wanted more control over what you do or don't hear? This talk explores the potential of Python, deep learning, and open databases to bring you towards that goal, without needing expensive licenses or software.
-
The Sound of Silence: Online Misogyny and How we Model it ▶ Recording 📝 Transcript
Teresa Ingram
Female-identifying people are being attacked and silenced online. Social media platforms act as neutral bodies and law enforcement can’t stop the abuse. When you can’t trust that your safety will be protected online,...
-
Time Series Anomaly Detection for Bottling Machine Maintenance
Andrea Spichtinger
This talk discusses time series anomaly detection methods for predictive maintenance of machines in bottling plants and their implemention on AWS edge devices.
-
Time series modelling with probabilistic programming
Sean Matthews, Jannes Quer
This talk focuses on a Bayesian approach of advanced time-series forecasting in the case of small data. We describe the stages of modelling from simple smoothing to advanced forecasting by applying Gaussian Processes...
-
Tools that help you get your experiments under control ▶ Recording 📝 Transcript
Katharina Rasch
There is now a wealth of tools that support data science best practices (e.g. tracking experiments, versioning data). Let’s take a look at which tools are available and which ones might be right for your project.
-
Transforming a Legacy System into a Bias-Mitigating AI Solution for Debt Repayment
Avaré Stewart
We present Phoenix, a modernization of a legacy, German rule-based system based on Tesseract and SpaCy, and uses AI Fairness 360 to build tunable, Bias-Minimizing AI solutions to offer payment incentives to debtors.
-
Using adversarial samples to break and robustify your Vision Neural Network Models
Irina Vidal Migallón
We will cover several techniques to expose weaknesses and robustify neural network models for computer vision, from basic precautions to more advanced adversarial training.
-
Using machine learning for Level Generation in Snake (video-game) ▶ Recording 📝 Transcript
Filipe Silva
As a practical example, this tutorial uses machine learning models to predict where to best place the apple in Snake. By using datasets that contain different plays we can obtain different game experiences or models...
-
Using Overhead Video Capture to Analyse Grouping Behaviour of Dancers in a Silent Disco
Nelson Mooren
I built upon Python's OpenCV library to detect locations of dancers in a silent disco, using their headphone lights as a proxy and performed network analysis to investigate their grouping behaviour based on the...
-
Version Control for Data Science ▶ Recording 📝 Transcript
Alessia Marcolini
Are you versioning your Machine Learning project as you would do in a traditional software project? How are you keeping track of changes in your datasets?
-
Visualizing Interactive Graph Networks in Python ▶ Recording 📝 Transcript
Jan-Benedikt Jagusch
In this talk you will learn how to visualize graph networks in Python, using `networkx`, `traitlets`, `ipywidgets` and `plotly`. The resulting plot will be fully interactive, which makes it easy to filter edges, find...
-
vtext: text processing in Rust with Python bindings ▶ Recording 📝 Transcript
Roman Yurchak
In this we talk present how to write Python extensions in Rust, and discusse advantages and limitation of such approach. We then illustrate this approach on the vtext project, that aims to be a high-performance...
-
Want to have a positive social impact as a data scientist? ▶ Recording 📝 Transcript
Ellen König
Discover your individual approach towards more positive social impact by conducting experiments. I'll show you how! I’ll also share my learnings from doing such experiments over the last 15 years.
-
Where Linguistics meets Natural Language Processing ▶ Recording 📝 Transcript
Mariana Capinel
This talk explains how linguistics describes language - via phonetics-phonology, morphology, syntax, semantics and pragmatics. We will combine linguistic concepts with models through examples for NLP newbies.
-
Why you don’t see many real-world applications of Reinforcement Learning.
Yurii Tolochko
Reinforcement learning is the closest thing to a general AI system that we have. When it works, that is. The problem is, it often doesn’t. In this talk we will discuss the difficulties of RL as well as what to look...
-
Why you should (not) train your own BERT model for different languages or domains
Marianne Stecklina
Language models like BERT can capture general language knowledge and transfer it to new data and tasks. However, applying a pre-trained BERT to non-English text has limitations. Is training from scratch a good (and...
-
Your Name Is Invalid! ▶ Recording 📝 Transcript
Miroslav Šedivý
About people with first names, middle names, last names, one-word names, multiple names, and changing names, about names with characters beyond ASCII, and about using Python to handle them correctly, because names of...