cpp11 - welding R and C++

Jim Hester

R and S have a long history of interacting with compiled languages. The Rcpp package has been a widely successful project, however over the years a number of issues and additional C++ features have arisen. Modifying Rcpp to fix or include these would require a great deal of work, or in some cases would be impossible without severely breaking backwards compatibility. cpp11 is a ground up rewrite of C++ bindings to R with different design trade-offs and features. This talk will focus on the new features of cpp11 including:

• Enforcing copy-on-write semantics.
• Improving the safety of using the R API from C++ code.
• Supporting ALTREP objects.
• Using UTF-8 strings everywhere.
• Applying newer C++11 features.

We will also discuss the mechanics of transitioning packages from Rcpp to cpp11 and why we feel cpp11 is a solid foundation to build upon in the years ahead.

R at Ohio Supercomputer Center

Wilbur Ouma

An introduction to the high-performance computing (HPC) clusters at Ohio Supercomputer Center (OSC), with an emphasis on using the RStudio server at OSC.

Three Levels of Computational Mobility

Cole Brokamp

Transferring an R project from one computing environment to another is a common task, including moving a local R project to a high performance computer, sending a project to a collaborator (including your future self), deploying a Shiny application, deploying R code to a continuous integration server, or creating a research compendium to ensure the reproducibility of your project. This process requires the capture and deployment of specific versions of R packages, their system dependencies, and R itself. In this talk, I will give an overview of thee different mechanisms for accomplishing this task: (1) using a static package repository (e.g., MRAN snapshots), (2) using helper packages (e.g., {dep}, {renv}, {sysreqs}), and (3) using containers (e.g. Docker, Singularity). The desired balance between reproducibility and complexity for a specific project should direct what might be the best method. I will discuss these tradeoffs and how these three mechanisms can work separately and together to achieve true computational mobility.

sARcASm: Practicing programming techniques via an R package

Alec Wong

Several social media communities have converged on intra-word randomized mixed casing as the accepted method of conveying sarcasm or mockery in printed text. The {sarcasm} package fills a critical gap in the R package ecosystem by making this mode of expression available for use in R programming, and thereby reporting and other important documentation.

Actually, {sarcasm} is a goofy exploration which began with an attempt to produce said intra-word randomized mixed casing in a functional fashion, and evolved into a small exploration of error handling functions. Replicating this behavior in Python served as additional exercise in Python programming, module creation, and the use of the {reticulate} package. The minimal dependencies and code also make it a good demonstration of R package infrastructure.

R packages are an excellent way to practice writing reproducible code, to test new techniques, and share your work – even if the purpose is silly.

sarcasm package on GitHub

R-aspberry Pi!

Matthew Hoff

How I deployed a simple machine learning model using R & Shiny installed on a Raspberry Pi.

Analyzing Bike Share Data in R

Trevin Flickinger

Public bike share has been available in Columbus since 2018. In this talk we will make visualizations, analyze the data using the Tidyverse, and build models to forecast bike sharing demand. We will also explore the introduction of electric bikes to the system in 2020.

Blog post on bike sharing

Bayesian Model Selection with the R Package slgf

Thomas Metzger

Linear models with categorical predictors are one of the most widely used statistical methods across a variety of scientific fields. However, common simplistic assumptions associated with such linear models may lead to model misspecification and ignore key relationships within the data. With the new R package ‘slgf’, I implement a model selection approach which utilizes Bayesian statistics to detect structures that would otherwise be disregarded by usual approaches. This R package allows the user to specify a set of candidate models, model structures, and Bayesian priors to quantify the posterior probability of each specified model. I illustrate this method and the usage of slgf through one-way analysis of variance (ANOVA), analysis of covariance (ANCOVA), and unreplicated two-way layouts using real data sets, although slgf is broadly applicable to the class of linear models that includes categorical predictors.

This session is based on joint work with Christopher Franck of Virginia Tech (Metzger & Franck, 2020).

Geomarker Assessment with R

Erika Rasnick

Relationships between place-based information and health outcomes are a common area of interest in medical and public health research. This talk will describe a set of publicly available tools and pipelines developed using R for transforming a residential mailing address to geographic coordinates and assigning a variety of place-based data (termed ‘geomarkers’). DeGAUSS, an offline container-based software powered by R, can be used for geocoding and estimating exposure to greenspace, distance to nearest roadway (a marker of air pollution exposure), access to medical care measured by drive time, and community-level poverty and deprivation, while protecting private health information. Further, we have developed a set of R packages that can be used for spatiotemporal geomarker assessment, including weather data and model-based estimation of traffic-related air pollution exposure.

Exploring Interactions in Regression Models with R and rFSA

Joshua Lambert

During the process of building a regression model, data scientists are sometimes tasked with assessing the effects of one or more variables of interest. With an additive regression model, effects (eg. treatment) are assumed to equal across all possible subgroups(eg. sex, race, age). This assumption is typically not true, and in the era of personalized medicine we must begin to demand more insight from our regression models. The package rFSA (Lambert, 2018), implements an algorithm meant to explore subgroup specific effects in large datasets by identifying 2 and 3 way interactions to add to multivariable regression models. This package supports the optimization of multiple criteria, and different modeling strategies. This talk will give a short introduction to rFSA and how it can be applied to regression models in R. A strategy for interaction visualization will also be discussed.

R For Communicating Science To Non-Scientists

Michael Sovic

Scientific and data literacies are arguably important characteristics of any citizenry, as they can underlie everything from personal lifestyle choices to the shaping of public policy. However, levels of such literacies vary with regards to conceptual knowledge and also the process by which scientific and quantitative understandings are constructed. Further, clearly communicating scientific information can be difficult even when the audience consists of domain experts, and so is especially challenging when the information is targeted to those outside scientific and data-driven communities. The importance of such literacy and effective communication has been apparent in issues ranging from vaccines to climate, but was highlighted again with the circulation of covid-19. Early in the pandemic, instances of miscommunications, misunderstandings, and misinformation led to several calls for expanded efforts by scientists to better engage with the public. As an attempt at one such effort, I began creating short videos that use scientific issues in the news as a springboard to talk about relevant data and concepts, and to also discuss the nature of scientific information and the process by which scientific understandings are built. In this talk, I’ll give examples of how R has provided me a valuable platform for creating visualizations to help communicate difficult concepts such as exponential growth, multiple testing effects, and reporting biases, to individuals outside of science.

R-Powered Excel

George Mount

Excel spreadsheets are the tool of choice for data analysis and reporting among many business professionals, but Excel has its limitations for building repeatable, dynamic data processes. In this session, users will learn how to begin automating Excel workbooks using the openxlsxpackage. This approach combines R’s knack for reproducible data analysis with the accessibility and end-user interaction features that make Excel so prevalent.

Tips for askeRs and helpeRs

PJ Van Camp

When asking others for help with our R code, it’s important to provide them with a Reproducible Example (ReprEx). This is the minimal data and code you need to recreate an issue. Using the ‘reprex’ and ‘datapasta’ package, this session will teach you how to create one. If you’re the expert helping someone else, it’s equally important you learn to structure your answer and provide not only a solution but also understanding. A few tips will be provided for them as well.

How We Used Rstats to Win Best Government Reporting in Ohio

Lucia Walinchus

In 2018, Eye on Ohio was one of just two U.S. outlets selected for a grant by the Pulitzer Center for Crisis Reporting to investigate property records.

A team of five people worked on this project for ten months, gathering, analyzing, and verifying millions of parcels. After interviewing dozens of people affected, writing approximately 700 lines of code to recalculate property values, and verifying figures with local officials, Eye on Ohio was able to put a price tag on a problem that affects communities in Ohio as well as most other states.

How Much are you Overpaying in Property Tax?

This project is based off of the simple idea that we can gain great insights by combining different property data sets. Ohio keeps great records, but unfortunately they are very county-centric. In other words, records tend to be in 88 different places! Which makes things difficult to say the least.

For the first story in this series, we looked at how entity sales affect property tax rates in Ohio. We know this has been going on for a while, but no one had ever actually quantified the damage. We know that if some people underpay their taxes, that means others overpay. But what does that actually COST our readers? That’s what we wanted to laser focus on.

For example, in Athens County, an apartment complex had been on the books for approximately $13.8 million. But then a mortgage appraisal of the property found it to be worth almost $49 million! However since the transaction was technically only a sale of the business, and not the building, the buyer only paid to have the name of the business changed. Total due to the county: 50 cents.

The story was picked up by 36 different news outlets, reaching an estimated half million people.

After local coverage, a bipartisan group of lawmakers addressed the issue in a new bill, and the Ohio Supreme Court issued an opinion that makes structuring such deals much harder.

The Cincinnati Society of Professional Journalists named it the Best Government News Story at their 2020 awards. I thought this would be a great way to show what R can do. And we published all our work on our website so others can learn from us.