March BARUG Meeting at Predictive Analytics World

Once again we are very happy to be able to hold our March meeting at Predictive Analytics World (PAW). As in previous meetings held in conjunction with PAW, BARUG members will be invited to the PAW reception before the meeting and will have access to the Exhibit Hall during the reception.

Also, BARUG members who wish to attend the PAW conference itself will be eligible for a 15% discount to the conference. The discount code is BARUGSF15.

Agenda:
5:30 to 7:00 PM - PAW reception
7:00 - Welcome and Announcements
7:05 - Eric Theise - Which Way is Inbound?
7:30 - Max Kuhn - Greatest Hits R Mixtape
7:55 - Nick Elprin - Deploying predictive models in R
8:20 - Antonio Piccolboni Randomized software testing in R with quickcheck (https://github.com/RevolutionAnalytics/quickcheck/blob/dev/docs/tutorial.md)

-------------------------------------------------

Eric Theise
Which Way is Inbound?

Though SF Muni routes are prefaced with the modifiers "Inbound" and "Outbound", their use of the terms often has little to do with their commonly understood meaning. In this presentation I'll use open data, an open source geostack, and R's circular package to visually, then statistically, analyze and discuss the resulting cognitive dissonance.

-------------------------------------------------

Max Kuhn
Greatest Hits R Mixtape

There are myriad features, functions or applications that could qualify as “great”. In the last few months, three that have brought a smile to my face. They are: the "three dots"/ellipsis, interactive graphics, and ggplot2’s grid.arrange. I’ll show examples of each along with why they qualify.

-------------------------------------------------

Nick Elprin
Deploying predictive models in R

We’ll demonstrate an easy way to “operationalize” your predictive R models by exposing them as low-latency web services that can be consumed by production applications. In the context of a real-world use case, we'll describe some of the more subtle requirements for hosting predictive models, including zero-downtime upgrades and retraining/redeploying against new data. Finally, we’ll describe some best practices for writing R code that will make your predictive models easier to deploy.

------------------------------------------------

Antonio Piccolboni
Randomized software testing in R with quickcheck

Dijkstra famously said that "Testing shows the presence, not the absence of bugs", but, with the availability of cheap computing power and interpreted languages, programming has evolved away from Dijkstra mathematical approach in favor of an experimental, trial and error one. Unlike Dijkstra, statisticians and contemporary computer scientists are much more comfortable with a probabilistic notion of correctness and testing has become a mandatory activity, but still an onerous one. In this talk we introduce the package quickcheck, whose inspiration is a package by the same name for the language Haskell. It supports a style of writing tests that requires formulating assertions about what functions should do and what the inputs should be, but doesn't require to write the actual inputs and outputs. Tests are more general, closer to specifications and provide more coverage with less testing code. Moreover, the thoroughness of testing can be increased changing a simple parameter and the developer is less likely to incorporate implicit assumptions in the choice of data points, or is forced to make them explicit. For example, to test that the transpose function is an involution for any matrix, we only need to write:

test(function(X) identical(X, t(t(X)), list(rmatrix))

This talk will provide you with the knowledge to incorporate quickcheck in your development activity right away and to approach the goal of writing "trustworthy software" set by John Chambers in his Prime Directive.