Iterating over Statistical Models: NCAA Edition.

13 Apr 2016

On Saturday, April 9, I spoke at the New York R Conference. The slides from my talk are available: 2016-04-ny-r-conference.pdf. Video was recorded and will be posted soon.

My Talk

Naturally, I talked about Stan. The point I wanted to get across was that statsitical modeling should be treated as a discipline. On the stan-users list and what I know to be common practice, I see people embedding statistical models within scripts. This makes it hard to collaborate and the software world has figured this out with tools like git.

We collaborate all the time. For the statistical models built for the Machine Madness Kaggle competition, we built models, checked them in, discussed them. If Rob Trangucci wasn’t around, we woudln’t have competed this year.

The Conference

Jared Lander, Jessica Lin, and the two crews from Lander Analytics and Work-Bench did a great job of organizing the conference. Each speaker was given 20 minutes to talk; no questions. It ran pretty smoothly and I picked up a lot of information about R’s development.

For me, highlights included:

• JJ’s talk on RStudio’s new features and bookdown.org
• Alp’s talk on ADVI
• Andrew’s talk on social penumbras
• Drew’s talk on social aspects of data science
• Vivian’s talk about evoking emotions through data
• Bas’s talk on program analysis
• Josh’s talk on building packages for use at New York Times

A recurring theme this year was testing, which is a big step in the right direction.

For a running commentary of the conference, see #rstatsnyc on twitter.

Questions

If you have questions about my talk, feel free to reach out.