Conference Report: Why R? 2019
Michał Burdukiewicz
Warsaw University of Technology, Why R? FoundationFilip Pietluch
University of WrocławJarosław Chilimoniuk
University of WrocławKatarzyna Sidorczuk
University of WrocławDominik Rafacz
Warsaw University of TechnologyLeon Eyrich Jessen
Technical University of DenmarkStefan Rödiger
Brandenburg University of Technology Cottbus–SenftenbergMarcin Kosiński
Gradient Metrics LLC, Why R? FoundationPiotr Wójcik
University of Warsaw, Data Science Lab2020-06-01
Abstract
The ‘Conference Report: Why R? 2019’ article from the 2020-1 issue.
Why R? 2019 conference
Why R? conferences have been the hallmark of the Why R? Foundation (whyr.pl). Our goal has been to establish a series of international R-related events in Poland. After three years, we are happy to announce that our main event, the Why R? conference, has become one of the largest annual R conferences in Central Europe.
Why R? 2019 was the third part of Why R? conference event. After the last edition that was held in Wrocław (Burdukiewicz et al. 2018), our conference has returned to Warsaw. A total of approximately 300 people from 20 countries attended the main conference event. The event took place from 26th to 29th September 2019 and was co-organised by the Faculty of Economic Sciences of the University of Warsaw (wne.uw.edu.pl/en/), a leading academic institution in Poland, having important achievements in quantitative methods and data science. We received major support from ML in PL Society (mlinpl.org), a group of young researchers, aiming to promote machine learning events in Poland, who shared their resources and experience to make the conference more accessible.
For the first time, this year the conference featured a language-agnostic data visualizations hackathon (whyr.pl/2019/hackathon). Such an event gives the Why R? community a chance to exchange experience and inspirations with the users of any other languages and tools.
Participants
In spite of the fact that Why R? events are aimed at experienced data science practitioners, each conference gathers a high percentage of students (around 30%). Our participants have very diverse scientific backgrounds, where mathematics (mainly statistics) and computer science are the most common. All of them have jobs related to data science, including professional R developers (programmers), data engineers, machine learning practitioners and business analysts. One of the key advantages of Why R? is that it gathers participants both from academia and the industry.
Conference program
The format of the conference was aimed at exposing participants to recent developments in the R language as well as a wide range of application examples. The event consisted of workshops, invited keynote talks, field-specific series of talks, lightning-talks, special interest groups and a full-day data visualizations hackathon. It offered extensive networking opportunities. The welcome party was held at the conference venue on the first day of lectures. In addition, many informal gatherings were organised during each conference day, as the event took place close to the Old Town.
To sum up, Why R? 2019 consisted of: one day of hackathon (60 attendees), one day of workshops (150 attendees), one evening of round tables, two days of lectures (250 attendees) and one evening Welcome paRty (100 attendees). In 2019 we hosted a total of 315 unique attendees. During lectures there were carried out: 6 keynote talks, 42 regular talks and 14 lightning talks. Below you can find the conference agenda.
Materials from the conference are available on GitHub and YouTube: - abstracts github.com/WhyR2019/abstracts, - presentations github.com/WhyR2019/presentations - videos whyr.pl/youtube/
Data Visualizations Hackathon
On the day before the conference we organized the free Data Visualizations Hackathon. It was a great opportunity for networking and exchange of experiences between data scientists that use different programming languages. The challenge was based on the data from Google Places API, which allows to search for places in a particular area. Thanks to this API we gathered data related to places in Warsaw, their working hours and occupancy. Based on this source of data participants, divided into 10 teams, were asked to prepare useful business application powered data visualizations solutions and techniques.
Pre-meetings
In 2019, Why R? 2019 was preceded by fourteen pre-meetings in eight countries. The purpose of those meetings was to provide the space for professional networking and knowledge exchange for practitioners and students, from the area of statistical machine learning, programming, optimization and data science. The Why R? Foundation supported organisation of pre-meetings financially and/or by sending speakers.
The organisation of pre-meetings would not be possible without the wonderful support of local R communities. Aside from the promotion of Why R? we had a great opportunity to interact with other R enthusiasts.
Workshops
Why R? 2019 conference had a wide portfolio of workshops that are listed below. One can find materials from workshops at this GitHub repository github.com/WhyR2019/workshops
- Introduction to modern Generalized Additive Models in R (with mgcv) by Matteo Fasiolo (University of Bristol). The assumption of the full-day workshop was firstly to give its participants some theoretical background about GAMs and some practical experience in R and finally to make attendees ready to start applying these models themselves. GAM models are a non-parametric extension of traditional regression model and were proved to be highly useful for both predictive and inferential purposes. Their popularity is based on a good balance between flexibility and interpretability as well as on the possible application on large datasets. Matteo started with explanation of standard GAMs and related R packages. He explained what an additive model is, how the smooth effects and random effects are introduced. GAM models fitting was accompanied by the explanation of additional diagnostic and model selection tools, and Big Data GAM methods. In the end more recent developments were also described, i.e. quantile GAM models. The practical sessions were based on the mgcv (Wood 2017), qgam (M. Fasiolo et al. 2017) and mgcViz (Matteo Fasiolo et al. 2018) packages.
- data.table introduction & time-series by Jan
Gorecki (H2O.AI). The workshop was divided into two parts – the first
part was devoted to the introduction of
data.tablequery concept while the second focused on a particular use case of working with time-series data. In the first part Jan showed syntax similarities and differences betweendata.tableanddata.frameapproaches. He used Arun Srinivasan workshops materials from useR!2017, with a few extras: chaining ofdata.tablequeries, reference semantics, subset of data:.SDand R function argument matching. In the second part Jan showed the application of efficient data processing on financial time series data of high-frequency (tick data quotations), including efficient aggregation to OHLC data, calculation of moving averages and using rolling join. - Straightforward introduction to Deep Learning in R (with Keras) by Mikołaj Bogucki and Mikołaj Olszewski (iDash). The workshop started with the explanation of what Deep Learning and Neural Networks are (complex functions) and what components they include (input, output, hidden layers and weights). Then Keras (Allaire et al. 2018) was presented as a high level library allowing to build neural networks with an easy to use set of commands. The practical example using airBnB data and Keras R codes showed all stages of building a Neural Network: (1) defining the structure of the network, (2) defining the way of training (the loss function and the optimizer algorithm), (3) training (together with its visualization), (4) evaluation and (5) prediction. In addition, training, validation, test set division, simple imputation of missing data, using non-linear activation functions and basic feature engineering was shortly explained.
- auditor + DALEX: a powerful duet for validation and explanation of machine learning models by Alicja Gosiewska and Tomasz Mikołajczyk (MI2 Data Lab). The aim of the workshop was to familiarize participants with modern methods of model verification and exploration. In the first part Alicja and Tomasz introduced the idea of DALEX (Biecek 2018) explainers, showing how to use them to assess the performance of a model and explain the model’s predictions (including global and local explanations). In the second part they focused on additional functionalities of the auditor (Gosiewska and Biecek 2018) package, showing how the analysis of residuals may be applied to select the best model or even improve models.
- Black is the new White - using eXplainable Artificial Intelligence in Business by Marcin Chlebus (Faculty of Economic Sciences, University of Warsaw, Data Juice Lab, Data Donuts). Marcin presented XAI as a possible solution for understanding “Black-box” model complexity and fuzziness. He showed how XAI helps in stability and sensitivity analysis, prediction quality assessment and identification of decision drivers. The use cases showing the application of XAI in cross-sell marketing campaigns and risk management were presented. With the use of step by step analysis, it was shown that XAI is a set of tools enabling application of “black box” models in many business industries through in-depth understanding of advanced machine learning modelling.
- Shiny Basics by Theo Roe (Jumping Rivers). This workshop was intended as a quick introduction to creating interactive visualisations of data using shiny. Theo started with some basic examples of using rmarkdown and htmlwidgets, then showed input and output bindings to interact with R data structures and using inputs to render output tables and graphs. In the end, Theo showed how to create own page layouts using shiny and shinydashboard and input and output "slots".
- Speeding up R wih C++ (Rcpp) – from basics to more advanced
applications by Piotr Wójcik (Faculty of Economic Sciences,
University of Warsaw, Data Science Lab). Piotr discussed various aspects
of Rcpp that helps to easily replace the R code with
often significantly faster counterparts in C++. Writing R functions in
C++ was explained, starting from simple examples with the focus on
similarities and differences between R and C++ syntax. Then Piotr at
first explained writing loops and recursive calls in C++, using Rcpp
sugar and secondly presented how to store C++ code in
*.cppfiles, using Standard Template Library, iterators, algorithms and range-based loops. In the end complex input/output objects (S3 and S4) were discussed. - Machine Learning Pipelines and Reproducible Research with
mlr3 and drake by Jakob Richter (TU Dortmund University) and
Patrick Schratz (LMU Munich). The workshop was divided into two parts –
the former introduced the new mlr3 package (Lang et al. 2019) framework (the successor of
the mlr package) while the latter presented a brief
overview of the drake package (Landau 2018) in R. In the first part Jakob and
Patrick explained the philosophy and ingredients of
mlr3 package. They presented how to define the data and
the target variable, using learners provided by mlr3,
set and tune hyperparameters, make predictions and evaluate their
performance, including resampling techniques and comparing multiple
learners. The practical example showed hyperparameter tuning and
training of a random forest classifier on the
irisdataset. The practical part also involved benchmark analysis of multiple learners, using different hyperparameter ranges on theirisandspamdatasets. A particular emphasis was put on machine learning workflows that might be easily controlled with mlr3pipelines package (Binder et al. 2019). In the second part the drake package was presented. It helps to set up a reproducible workflow of the project and it easily integrates with the mlr3 package and its extensions. - Basics of spatial data analysis by Jakub Nowosad (Adam Mickiewicz University, Poznan). The emphasis in this workshop was put on getting started with spatial data analysis. Jakub demonstrated key packages for spatial analysis and making maps, explained spatial data representation in R, using sf (Pebesma 2018), for spatial vector data, and raster (Hijmans et al. 2017) packages. Then he gave a lot of examples of spatial data visualization, using a powerful tmap package (Tennekes 2018), including some vector-raster interactions. In the end, he also showed data manipulation examples with the tidyverse (dplyr) approach, in which sf spatial objects are simply special data frames.
Invited talks
The invited talks topics included domain knowledge from statistics, computer science, natural sciences and economics. The speakers list presents as follows:
Marvin Wright
Random forests used to be everywhere, from Microsoft Kinect to meteorology, but their popularity considerably dropped with the advent of deep learning. During his keynote talk at Why R? 2019 Marvin R. Wright has shown that random forests still can be used in machine learning routines, making the whole process time- and cost-efficient.
Implementing a real-life machine learning solution is not only about the best performance. Marvin has shown that considering trade-off between performance and costs of the analysis, random forests are still unbeatable. Aside from the methodological background, Marvin has given an overview of random forest implementations in R (Wright and Ziegler 2017).
Marvin is a Postdoc at the Leibniz Institute for Prevention Research and Epidemiology in Bremen, Germany. He is the author of several R packages, including the fastest implementation of random forest in R, ranger. He holds a Ph.D. in Biostatistics from the University of Lübeck, supervised by Andreas Ziegler. In the past, Marvin worked at the University of Lübeck. He was a visiting researcher at the University of Copenhagen. Also, he spent some time in the automotive and health insurance industries. His main research interests are interpretable machine learning, genetic epidemiology and survival analysis.
Jakub Nowosad
Jakub Nowosad’s keynote lecture was a great opportunity to learn about geostatistics. Jakub, a co-author of the Geocomputation with R (Lovelace, Nowosad, and Muenchow 2019), has focused on tools used to solve real-life problems in spatial data analysis.
The growing importance of spatial data stimulates a rapid evolution of geostatistical methods. Jakub, as the active member of #rspatial community, not only presented cutting-edge tools but also gave his unique insight into the future of the spatial data analysis.
Jakub is an assistant professor in the Department of Geoinformation at the Adam Mickiewicz University in Poznan, Poland. His main research is focused on developing and applying spatial methods in order to expand our understanding of processes and patterns in the environment. He has extensive teaching experience in the fields of spatial analysis, geostatistics, statistics, and machine learning.
Sigrid Keydana
We know how accurate are our predictions but do we really know how certain they are? This question has been answered by Sigrid Keydana (RStudio) during her keynote lecture.
Sigrid has presented tfprobability, an interface to TensorFlow Probability, a tool for obtaining uncertainty estimates from deep neural networks. This exciting tool can be extended beyond a classic deep learning framework into complex hierarchical models.
Sigrid is an Applied Researcher at RStudio. She has experience as a psychologist, software developer and data scientist. She is passionate about exploring the borders of deep learning, especially by helping users to apply the power of deep learning in R.
Steph Locke
Machine learning models find their place in almost every area of our life, influencing things as small as the video recommendations on YouTube or as big as the length and severity of a sentence in a criminal procedure. With the growing importance of machine learning, it becomes more and more important to train models while keeping in mind their ethical consequences.
During her keynote talk at Why R? 2019, Steph Locke showed us ethical concerns about data science. Apart from pointing out existing issues, she has also presented solutions leading to more fair and transparent machine learning models.
Steph is the founder of a consultancy in the UK. Her talks, blog posts, conferences, and business all have one thing in common – they help people get started with data science. Steph holds the Microsoft MVP award for her community contributions. In her spare time, Steph plays board games with her husband and takes copious pictures of her doggos.
Wit Jakuczun
Wit Jakuczun from WLOG Solutions presented his talk about deploying - How to make R great for machine learning in (not only) Enterprise.
For many years software engineers have put enormous effort to develop best practices to deliver stable and maintainable software. How R users can benefit from this experience? Wit answered this question by going through several concepts and tools that are natural for software engineers but are often undervalued by R users.
Paula Brito
During her keynote lecture at Why R? 2019 Paula Brito has given a unique insight into the world of symbolic data, where data points are represented not as single values, but more complex structures, like sets or intervals (Noirhomme‐Fraiture and Brito 2011).
A classical paradigm of data science assumes that categorical variables, like gender or educational stage, are represented as the single value per observation. Paula has shown how to utilize her package, MAINT.Data, to model interval data, using its symbolic representation which leads to more accurate and robust models.
Paula is Associate Professor at the Faculty of Economics of the University of Porto, and member of the Artificial Intelligence and Decision Support Research Group (LIAAD) of INESC TEC, Portugal. Her current research focuses on the analysis of multidimensional complex data, known as symbolic data, for which she develops statistical approaches and multivariate analysis methodologies.
Round tables
Round tables are networking-oriented social mixers devoted to connecting people with similar interests. The exact points discussed during the round table and its style depend on the moderators who are shaping out the details, based on the general agenda provided by the Why R? organizers. The organizing committee both selects the topics of round tables and invites appropriate moderators.
Diversity in Data Science
This board aims to inspire members of affinity groups to pursue careers in data science. We hope that this platform for networking will reduce the diversity of R community. Moderator: Barbara Sobkowiak (Women in Machine Learning & Data Science Poland).
Career-planning in Data Science
Participants of WhyR will have a chance to learn from more experienced R enthusiasts about their career paths. Moderator: Kamil Kosiński (PwC).
Teaching Data Science
Practitioners will share their experiences in introducing their students to basic and advanced concepts of data science. Moderator: Patrick Schratz (Ludwig Maximilian University of Munich).
Data Visualizations
Discuss data visualizations good practices and approaches to various presentation challenges. Moderator: Michał Burdukiewicz (Warsaw University of Technology).
Ethics in Data Science
With the increased importance of machine learning, we are becoming more and more concerned about the ethics of data science. Moderator: Steph Locke (Locke Data).
Conference organizers
The organizing committee consisted of Klaudia Korniluk, Marcin Kosiński, Michał Burdukiewicz, Jarosław Chilimoniuk, Katarzyna Sidorczuk, Filip Pietluch, Weronika Puchała and Dominik Rafacz.
The quality of the scientific program of the conference was the achievement of Stefan (Brandenburg University of Technology Cottbus-Senftenberg), Piotr Wójcik (University of Warsaw) and Bernd Bischl (Ludwig Maximilian University of Munich).
Acknowledgements
We would like to express our gratitude to all our sponsors, the Faculty of Economic Sciences (University of Warsaw), ML in PL Society, local organizers of the pre-meetings and student helpers.
Additional information
Why R? 2019 website http://whyr.pl/2019 Corporate sponsors: PwC Poland, iDash, R Consortium, umping Rivers Ltd., Appsilon Data Science, RStudio, Inc., AnalyxGmbH, Pearson IOKI and WLOG Solutions.