Home » Security Bloggers Network » Quick hit: Some ggplot2 Stat

Quick hit: Some ggplot2 Stat

by hrbrmstr on June 18, 2019

I’ve become a big fan of the {logspline} package over the past ~6 months and decided to wrap up a manual ggplot2 plotting process (well, it was at least in an RStudio snippet) into a small {ggplot2} Stat to make it easier to visualize various components of the fitted model.

If you’re new to logspline density estimation this paper by Stone, Hansen, Kooperberg, and Truong is a pretty accessible introduction to the methodology and the (primary) algorithm used in the {logspline} package.

If you’re already a user of the {logspline} package have have either been just using the package’s base S3 plot function but wanted the functionality in {ggplot2} this was pretty much designed for you.

If you hit up https://git.rud.is/hrbrmstr/gglogspline there are links to all your fav social code collaboration services plus installation instructions. There’s a single Stat — stat_logspline() — which does all the hard work. By default it will make y the stat(density) computed field but there are other computed fields you can use as well:

count: computed count estimates (similar to ggplot2::stat_density())
probs: distribution function
survival: survival function
hazard : hazard function

You can take a look at the (light) comparison between geom_histogram(), geom_density(), ggalt::geom_bkde(), and gglogspline::stat_logspline() below (or use this link to bust the iframe):

Sponsorships Available

FIN

For “general purpose” density estimation you can likely still rely on stat_density() but after you dig into the background of logspline and try it out in some appropriate use-cases you may grow to like it as much as I have.

The package itself is super-small and focused. One reason for that is it may be helpful for those who want to create or customize their own ggplot2 Stats. (I think I forgot to change the license to MIT before this post so I’ll rectify that ASAP; I default to AGPL in a fairly pathetic attempt to stop skeezy cybersecurity startups — pretty much 99% of them — from stealing code without giving back to the community).

As usual, kick the tyres and file issues or PRs as you feel moved (and wherever you feel comfortable social coding).