Arben Kqiku

Overfitting & overspending, a digital marketing tale

Nowadays, the great majority of digital advertising campaigns are powered by machine learning algorithms. Advertising platforms such as Google Ads encourage users to exploit its ML algorithms from the onset. However, ML algorithms require data to function properly. As a consequence, during the initial learning phase advertising platforms will spend a lot of money to understand what works and what doesn’t. This process doesn’t necessarily bring the expected results, which leads to advertisers being disappointed. In my talk, I will explain how we can reap the benefits of machine learning without falling prey to overspending.

Benjamin Abt

Data Scientist at PartnerRe with an actuarial background. Develops R based tools for insurance and provide R-trainings within the company.

Using R in Production at PartnerRe: Framework and applications

The output of PartnerRe’s data science team is often deployed to production. Examples include automated reports used by decision makers, tools for reinsurance pricing (e.g. simulation engines) and trained machine learning models that support work flows in our company. In our context, production’ means that something is used by non-technical users in a business critical context. Consequently, production content needs to be highly available and users expect complete reproducibility of all results. To make R fit for this purpose, we have developed a set of best practices around how we structure and version control our code and the environments our code runs in. In this talk, we focus on the open source infrastructure supporting this process: version control, continuous integration, containerization and container deployment and orchestration. We conclude this presentation by illustrating this framework through some applications.

Daniel Piqué

Daniel is 7th year MD-PhD candidate at Albert Einstein College of Medicine (Einstein) in the Bronx, NY. He completed his PhD in the Department of Systems and Computational Biology at Einstein in 2018 with Drs. Jessica Mar and John Greally where he developed two R packages (oncomix, receptLoss) and a RShiny web application (aneuvis) for analyzing and visualizing genomic data. He is currently in his final months of medical school and will start his clinical residency in July 2020.

A lesson in biomedical research reproducibility: How I discovered a missing data point in a paper with 8000+ citations

Reproducibility of foundational, highly-cited studies in biomedical research is important but technically difficult, as data sets are often not easily accessible. This is particularly true for important papers published before the dawn of modern computing. Here, I attempt to reproduce Figure 1 from a paper published in 1971 that has received 8000+ citations and formed the basis for the two-hit hypothesis, a fundamental theory in cancer biology. The data used to produce Figure 1 was provided in the paper as Table 1, which I converted into a R data package called ‘knudson.’ Using the data from Table 1, along with tools from the tidyverse, Figure 1 could be replicated nearly identically, with the exception of 1 missing data point and error in the exponent of the fitted curve. The lessons learned from attempting to reproduce Figure 1 from the data in Table 1 are contained as a vignette within the ‘knudson’ package (available on github) and will be discussed in this talk. I will conclude by discussing strategies for ensuring that highly-cited research papers, both past and present, are reproducible.

David Granjon

At Novartis, David provides support to develop production Shiny apps, HTML templates and train other associates in package development. David also contributes to the R Shiny ecosystem through its open source project, namely RinteRface, bs4Dash (Bootstrap 4 shinydashboard), shinyMobile, the virtual physiology simulator.

shinyMobile: Mobile-Ready Shiny Apps

Have you ever dreamt of a template allowing you to build native looking IOS/Android Shiny apps? For two years, RinteRface soldiers strived to deliver a robust solution aiming at bridging the gap between desktop shiny apps and mobile ready user interfaces: welcome shinyMobile (https://github.com/RinteRface/shinyMobile).

Duc-Quang Nguyen

Duc is a Data Scientist at Tamedia Romandie (Tribune de Genève, 24 heures, le Matin Dimanche). Previously, Duc worked as a data journalist at SRG-SSR for swissinfo.ch, quantitative analyst for a hedge fund and did a PhD in computational biology.

R in the newsroom

From breaking news to long term investigations, R is amply used in data journalism. It is present in every step: for scraping and wrangling data, for EDA and even for publishable graphics. I will walk through how I use R by profiling some published stories.

Emma Jablonski

Emma Jablonski is part of a team working on a platform called Renku for reproducible and collaborative (data) science research at the Swiss Data Science Center (a joint center between EPFL and ETHZ).

Reproducible data science with the RENKU platform

Communities and funding sources are increasingly demanding reproducibility in scientific work. We present RENKU (https://renkulab.io/): an open-source platform for reproducible data science. From an end-user perspective, through a unique authentication mechanism, the platform provides a seamless integration of an RStudio Server or a Jupyter notebook, git version control, git LFS to handle data, continuous integration via GitLab, containerization via Docker images that can be reused and shared, and automatic workflow generation via a knowledge graph. With RENKU, every step of the data science research that generates new code or data is preserved. This allows scientists to step backwards through the history of their research and retrieve earlier versions of their methods and results. RENKU materializes data science recipes and data lineage into a knowledge representation based on the Common Workflow Language standards (CWL), and PROV-O ontology. Data lineage is automatically recorded and workflow captured within and across RENKU projects, allowing derived data and results to be unambiguously traced back to original raw data sources through all intermediate processing steps.

Eric Röthlisberger

Eric works on education data flows. He appreciates collaborating with IT team at the start of the flow and with pedagogues for the construction of indicators. Current interest: sequence analysis to understand life courses. He has been working with R, SAS and Tableau for several years.

Studing life courses with sequence analysis

Our presentation is about analysing the life course of young people on their integration into the labour market. We use Sequence Analysis method with the package TraMiner. Our presentation will be about a complet analysis from a data modeling to result interpretation.

Giulia Ruggeri

Cartography in R: an overview of the geospatial visualisation landscape

Over the past years, R has become a comprehensive toolbox for visualizing geospatial data and perform geostatistical analysis. In particular, the {sf} and {raster} packages include all the necessary functions to visualize and manipulate vector and raster data. The visualization possibilities have although increased with packages such as {rayshader}, {leaflet} and {mapview}. In this talk we will see how to create maps in R, using different packages, going from static representation to interactive ones.

John Coene

John Coene is a Data Analyst at the World Economic Forum’s Strategic Intelligence team. John maintains to numerous open-source R packages such as waiter, echarts4r, and grapher. John is also part of the RinteRface Group where he contributes to packages such as shinybulma and fullPage.

Shiny user feedback with waiter

Shiny applications tend to lack user feedback, upon clicking on a button or selecting an item from a dropdown the user is left wondering whether their action actually registered. The waiter package (https://waiter.john-coene.com/) remedies to this by providing a way for the developer to programmatically show and hide loading screens with spinners or loading bars either on the entire app or on parts of the app. It’s extremely easy to use and allows making apps feel slicker.

Jonas Pipoz

Economist and project manager at the Neuchâtel Statistical Office with about 15 years of programming experience.

Studing life courses with sequence analysis

Our presentation is about analysing the life course of young people on their integration into the labour market. We use Sequence Analysis method with the package TraMiner. Our presentation will be about a complet analysis from a data modeling to result interpretation.

Kirill Müller

Kirill has worked on the boundary between data and computer science with more than 20 years of software engineering experience. He has enjoyed working with and contributing to the R ecosystem since 2012.

Relational Data Models with the {dm} Package

Kirill will present {dm}, a new package that facilitates working with multiple tables: data frames or database tables. Use cases include consuming, cleaning, defining and publishing relational data models.

Linda Maria de Cave

Linda holds a PhD in Mathematics and, after two postdocs positions, decided to move to Data Science. She is currently working as Data Scientist for Knowledge Lab AG. She is mainly focused on the development of highly tuned machine learning models, also by exploiting topological attributes of the datasets.

Topological data analysis

We showcase Topological Data Analysis (TDA) and related tools. TDA is a method for extracting patterns from data, which allows for powerful conclusions regarding not only data structures, but also relationships between them. TDA has been applied recently in genome research, social sciences, in the biomedical sector, particle physics and the financial sector. In our talk, we will give a small conceptual introduction to TDA, followed by a tutorial-style use-case (we will make material available on Github).

Marco Wirthlin

Marco, a cognitive scientist, is developing and applying Bayesian and causal models to human behaviour in the Fintech sector, where he works as a Data Science Consultant for Knowledge Lab AG. His main focus is understanding customer behaviour via interpretable, generative methods.

Topological data analysis

We showcase Topological Data Analysis (TDA) and related tools. TDA is a method for extracting patterns from data, which allows for powerful conclusions regarding not only data structures, but also relationships between them. TDA has been applied recently in genome research, social sciences, in the biomedical sector, particle physics and the financial sector. In our talk, we will give a small conceptual introduction to TDA, followed by a tutorial-style use-case (we will make material available on Github).

Mark Padgham

Mark recently joined rOpenSci as a Software Research Scientist, and also develops software for the World Health Organization to plan for encourage more sustainable and active transport.

rOpenSci, peer review, statistical software, and R testing

rOpenSci is an organisation dedicated to Transforming science through open data and software, primarily via our own system for peer-reviewing and curating R packages. Our peer review system accepts packages from a variety of categories mainly focussed on data workflows, and we have recently embarked upon a project to expand the scope of peer review to include statistical software–arguably the true core business of the R language. This project provides an opportunity to reconstruct and reconfigure how peer review of software might function. In this talk, I will briefly outline the future vision of this project, and then focus in particular on visions for how testing might be reconceptualised to play a more central role in the development and assessment of R packages.

Natasa Tagasovska

Natasa is a computer scientist by training and is currently finishing her PhD at HEC Lausanne. Her main research interest is a deeper understanding of learning models, hence works focuses on causal inference, generative models, uncertainty and interpretability.

Causal discovery from observational data in R - algorithmic information approach

Machine learning permeates our lives, with prospects to drive our cars and decide on our medical treatments. These ambitions will not materialize if the learning models remain unable to explain their behaviour when performing under diverse situations. Hence, any domain of science and application benefits from causal inference methods. While the golden standard for causal discovery is randomized control trials, experiments or interventions in a system are often prohibitively expensive, unethical, or, in many cases, impossible. In this context, an alternative is to use observational data to infer causal relationships. The objective of this talk is to tackle this challenge, by presenting three algorithms to distinguish causes from effects based on the independence of cause and mechanism postulate [1]). We will present the common intuition shared by three different approaches, their mathematical formulation and the corresponding R implementations and evaluation. [1] Bernhard Scholkopf, Dominik Janzing, Jonas Peters, Eleni Sgouritsa, Kun Zhang, and JorisMooij. On causal and anticausal learning. In ICML 29, pages 1255–1262, 2012.

Stella Bollmann

Stella has a background in Psychology and Statistics and has been working in Academia for many years as data scientist, statistical consultant and lecturer. She has a broad insight into the usage of R in different fields such as Statistics, Psychology, Education and Health Care.

Making R accessible for skeptics

As psychologist and statistician, Dr. Bollmann has a good understanding of two very different approaches to using R: The pragmatic approach on the one hand and the technical one on the other hand. In this talk, she will demonstrate how people, who do not consider themselves as mathematicians or computer scientists, can make themselves comfortable with R more easily to ultimately uncover its potential with regard to data management and analysis. Specifically, Dr. Bollmann will highlight (1) the benefits of R in contrast to generic software packages, such as SPSS, (2) readily available tools for data management and analysis in the form of specific R packages, and (3) potential challenges with regard to teaching the usage of R.