Reflections on UseR! 2013

This week I’ve been at the R Users con­fer­ence in Albacete, Spain. These con­fer­ences are a lit­tle unusual in that they are not really about research, unlike most con­fer­ences I attend. They pro­vide a place for peo­ple to dis­cuss and exchange ideas on how R can be used.

Here are some thoughts and high­lights of the con­fer­ence, in no par­tic­u­lar order.

  1. Håvard Rue spoke on Bayesian com­put­ing with INLA and the R-​​INLA pack­age. I was unaware of INLA before, but it is a much faster way than MCMC to do some Bayesian com­pu­ta­tions. It looks use­ful — I might try it sometime.
  2. Christoph Bergmeir (who has just fin­ished vis­it­ing me at Monash for a few months) talked about the Rsio­pred pack­age (not yet on CRAN) which uses a fuzzy mul­ti­cri­te­ria approach to fore­cast­ing ETS mod­els. Essen­tially, it tries to opti­mize RMSE, MAE and MAPE simul­ta­ne­ously, which will give biased fore­casts of course, but hope­fully more robust fore­casts. The opti­miza­tion is also bet­ter (in the sense of get­ting closer to the global opti­mum) than the ets() func­tion in the fore­cast pack­age. Christoph is also respon­si­ble for the big improve­ment in speed of the ets() func­tion from v4.05 of the fore­cast package.
  3. José Manuel Benítez Sánchez (Christoph’s boss) talked about the efforts of his team at the Uni­ver­sity of Granada to add machine learn­ing tools to CRAN.
    Their RSNNS pack­age looks good. Next time I fit a neural net, I’ll try it out.
  4. Dun­can Mur­doch gave an inter­est­ing talk on the new fea­tures in R 3.0.x and beyond. The most inter­est­ing part was that future releases will include the bug fixes and per­for­mance enhance­ments iden­ti­fied by Rad­ford Neal. In ques­tion time, Dun­can explained why we will prob­a­bly never have pack­ages depen­dent on spe­cific ver­sions of other packages.
  5. Steve Scott from Google talked about Bayesian struc­tural time series mod­els with regres­sors. Actu­ally, I’d heard his coau­thor (and boss) Hal Var­ian speak on the same sub­ject at the Oper­a­tions Research con­fer­ence in Rome last week. Look out for the bsts pack­age when it is released on CRAN. It looks very useful.
  6. As usual, there were lots of peo­ple talk­ing about Sweave and knitR for repro­ducible research. I quite like knitR because it uses mark­down which is a very sim­ple doc­u­ment markup lan­guage. How­ever, for my pur­poses, I still pre­fer keep­ing the tex and R files sep­a­rate as explained here.
  7. I heard two nice talks on visu­al­iz­ing lik­ert scale data by Kim­berly Speer­schnei­der (on the lik­ert pack­age not yet on CRAN) and Richard Heiberger (on the HH pack­age). Lik­ert scale data are the stan­dard fare of sur­veys, so it is good to see some seri­ous think­ing being done on how to graph the data usefully.
  8. I had my first expe­ri­ence of light­ning talks, where each per­son gets 5 min­utes. These were sur­pris­ingly effec­tive and inter­est­ing. I espe­cially enjoyed Andy South on map­ping half a mil­lion sig­na­tures.
  9. I gave a talk on the hts pack­age enti­tled R tools for hier­ar­chi­cal time series. I am now work­ing actively on hier­ar­chi­cal time series fore­cast­ing again, after a break from it for a cou­ple of years. So expect to see more on this topic in the com­ing months. It has a lot of appli­ca­tions, and there is a lot still to be done to develop the the­ory, method­ol­ogy and tools for han­dling hier­ar­chi­cal time series in practice.
  10. RStu­dio demon­strated their new debug­ging fea­tures to a few of us dur­ing one lunch break. In fact, they allowed us to down­load the pre-​​release ver­sion for our own use, but I’m not allowed to tell you the URL! How­ever, if you google [rstu­dio pre­view release] you might find it. The debug­ging fea­tures are excel­lent, so look out for v0.98 avail­able soon.
  11. I finally met the team from Rev­o­lu­tion Ana­lyt­ics, the peo­ple who pro­duce Rev­o­lu­tion R. Unfor­tu­nately, Rev­o­lu­tion R is not avail­able for 64-​​bit Ubuntu which is the plat­form I use.
  12. I was amazed at how much effort is going into mak­ing R work with gigan­ti­cally enor­mous and humungous data sets. For me a seri­ously big data set has a few hun­dred thou­sand obser­va­tions, but many peo­ple are work­ing with ter­abytes, petabytes and even exabytes of data. Mind-​​boggling.
  13. I can­not get used to the Span­ish habit of eat­ing in the very late evening, while still get­ting up for the first ses­sion at 9am. The con­fer­ence din­ner began at 9.30pm, and I left just after 12.30am as I wanted to get some sleep. The party was appar­ently still going well after 2am, includ­ing Steve Scott who was doing the 9am talk the fol­low­ing morning!
  14. This the most tweeted con­fer­ence I’ve attended. See the stream.
  15. I got to meet sev­eral peo­ple I’ve known elec­tron­i­cally for years, but never met in per­son includ­ing Colin Gille­spie (who was part of the first mod­er­a­tor team on cross​val​i​dated​.com) and Tal Galili (respon­si­ble for main­tain­ing R-​​bloggers).

Thanks to the com­mit­tee and spon­sors for mak­ing it a great con­fer­ence. The next one is in Cal­i­for­nia: UseR! 2014.


Related Posts:


  • http://robjhyndman.com/ Rob J Hyndman

    Thanks Andy. I’ve updated the link in the post. I didn’t see any­one record­ing the talk unfortunately.