Mark A. Wolters
Package scdensity unifies a few methods for enforcing shape constraints on probability density estimates. Because it uses the familiar kernel density estimator, getting shape-constrained estimates should be no more difficult than getting regular estimates.
Shape constraints are especially useful if you have a small data set or the data is noisy. For example, the
chickwts data set in R gives the weights of chicks that were fed with different diets. The
soybean groups have 12 and 14 observations, respectively. The figures below show what you get with a standard kernel density estimator (KDE) versus a shape-constrained KDE. The shape constraint used here was the
twoInflections+ constraint in
scdensity. It ensures that the estimate has only two inflection points, and its derivative has only three inflection points. The constraint makes the nonparametric estimate look more like a parametric one. Which option would you rather use to communicate your results?
The ffslides (“freeform slides”) LaTeX class started as a bunch of macros I wrote to allow me to create presentations using the article class. But as I have ended up using them exclusively to produce my presentations and even certain research notes for my own use, I thought it might be worthwhile to work it up into a document class. The class might be for you if:
If you’re interested, you can view the CTAN page, or you could download the manual (which is itself made using the class) to see what it’s about.
I have written a small R package called kofnGA. It implements a genetic algorithm to do subset selection. It’s for picking subsets of a fixed size k from n alternatives. The main selling point of the package is that it works with an objective function that is arbitrary and user-defined. So it should be worth a shot for any subset selection problem you happen to have. A paper demonstrating and evaluating the package is available in Journal of Statistical Software.
(If you’re curious about the little badge above with the download stats, see here)
There is sample code to accompany a chapter written by me and Charmaine Dean in the book Current Air Quality Issues. See the page.