user2016

Bell Labs in the 1970s was a hotbed of research in computing, statistics and many other fields. The conditions there encouraged the growth of the S language and influenced its content. The 40th anniversary of S is an appropriate time to relate a personal view of that scene and reflect on why S (and R) turned out as it did.

The speaker will discuss what he considers to be the most important outcome of his work developing TeX in the 1980s, namely the accidental discovery of a new approach to programming --- which caused a radical change in his own coding style. Ever since then, he has aimed to write programs for human beings (not computers) to read. The result is that the programs have fewer mistakes, they are easier to modify and maintain, and they can indeed be understood by human beings. This facilitates reproducible research, among other things.

I announced ggvis in 2014, but there has been little progress on it since. In this talk, I'll tell you a little bit about what I've been working on instead (data ingest, purrr, multiple models, ...) and tell you my plans for the future of ggvis. The goal is for 2016 to be the year of ggvis, and I'm going to be putting a lot of time into ggvis until it's a clear replacement for ggplot2. I'll talk about some of the new packages that will make this possible (including ggstat, ggeom, and gglayout), and how this work is also going to improve ggplot2.

We consider the problem of fitting a regression model that is both flexible and interpretable. We propose two procedures for this task: the Fused Lasso Additive Model (FLAM), which is an additive model of piecewise constant fits; and Convex Regression with Interpretable Sharp Partitions (CRISP), which extends FLAM to allow for non-additivity. Both FLAM and CRISP are the solutions to convex optimization problems that can be efficiently solved. We show that FLAM and CRISP outperform competitors, such as sparse additive models (Ravikumar et al, 2009), CART (Breiman et al, 1984), and thin plate splines (Duchon, 1977), in a range of settings. We propose unbiased estimators for the degrees of freedom of FLAM and CRISP, which allow us to characterize their complexity.

This is joint work with Ashley Petersen and Noah Simon at University of Washington.

The intuition and experience needed for sound statistics practice can be hard to learn, and a course that combines computing, statistics, and working with data offers an excellent learning environment in this regard. Moreover, an integrated approach to data science creates opportunities to reinforce statistical thinking skills throughout the full data analysis cycle, from data acquisition and cleaning to data organization and analysis to communicating results. As a result, students gain the ability to reason computationally, actively engage in statistical problem solving, and learn how to keep abreast of new technologies as they evolve. This talk describes approaches and provides examples for teaching data science in this integrated fashion.

Analyzing Big Data in real life poses challenges with respect to performance, methodology and reusability. R is well known for its succinct syntax for analytic tasks as well as plethora of tools for data analysis and visualization, but it is not always associated with scalability. In this talk we will present a scalable environment that allows the use of R (and other languages) in a collaborative setting, enabling sharing, reusability and reproducibility. In addition, it opens new possibilities for visualization and interactive graphics by providing seamless integration of JavaScript and R. Finally, the distributed nature of the design allows us to provide R tools that allow out-of-core data processing interfacing different back-ends including Hadoop without sacrificing the ease of use of R. We will also show a flexible framework for developing distributed models in R while re-using as much of existing work as possible. As part of this talk we will illustrate the use of those tools on real data sets, including interactive visualization and distributed computing.