At the end of august 2013, I was lucky enough to attend the Rencontres des jeunes statisticiens (young statisticians meeting). It is organised every other year by the Société Française de Statistique (French Statistical Society). It was thrilling to meet all these fellow statisticians in the making!
A lot of the talks were very interesting. In particular, one made me think of a recent post from Roger Peng on the famous Simply Statistics blog. Benjamin Guedj developed a nonlinear aggregation strategy, an approach aggregating different solutions (from different modeling) to a regression problem. It is implemented as an R package, COBRA, and it seems to performs quite good (and fast). Even though I suspect it might choke a little on difficult datasets (n<<p anyone?), I find it quite clever. And, relating to R. Peng post, it has the advantage of reducing the “researcher degrees of freedom”. Could it be a first step towards a deterministic statistical machine?