Iterating over Statistical Models: NCAA Edition.


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
  • 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.


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

continue / comment

Rent Party. 3/25.

Rent Party. 3/25. Broadway Dive

It’s been a while since I’ve had a gig. Been a little preoccupied with Stan, moving, a renovation, and a whole slew of other things. DJ Dalkesh asked me to do a guest spot at his monthly Rent Party at Broadway Dive on the Upper West Side.

Funk, soul, and hip hop all night. (good thing too… I don’t have the chops any more to do anything else.) Come through. It’ll be fun.

continue / comment

Map of Nu Delta Alumni - Part 2

I found Plotly’s capability to be a bit lacking, so I spend another couple hours updating the map. Since creating the original data set, more alums have chimed in, so they’ve been included.

I have a new version written in Leaflet for R. This package is just as easy to use as Plotly, but with more options.

Peter Lee (2013) pointed me to CartoDB and I got my first proptype working there in 5 minutes, but it didn’t have enough control. It looks like the underlying Leaflet package drives CartoDB.

Nu Delts

Ideally, there would be a slider for years. For now, each year is in its own group.

continue / comment

Map of Nu Delta Alumni

Recently, there’s been an email thread with Nu Delta alumni going back to 1956 checking in from different geographic locations.

I took an hour, grabbed the data, and put people on a map. (It took me longer to enter the data and figure out how to embed the image than the actual work.)

Nu Delts

Sorry the map is a little clunky. This my first time using Plotly.

Edit: Looks like there are some bugs in Plotly showing the text. It works with Firefox. I’ll look into a different plotting package when I have some time.

continue / comment