# Tidy and geospatial kernel mean shift clustering Tidy and geospatial versions of a kernel mean shift clustering for 1- and 2-dimensional data. ```r tidy_kms(data, ...) st_kms(x, ...) ``` ## Arguments - `data`: data frame/tibble of data values - `x`: sf object with point geometry - `...`: other parameters in `ks::kms` function ## Returns The output from `*_kms` have the same structure as the kernel density estimate from `*_kde`, except that `x,y` indicate the data points rather than the grid points, and `estimate` indicates the mean shift cluster label of the data points, rather than the density values. ## Details Mean shift clustering is a generalisation of $k$-means clustering (aka unsupervised learning) which allows for non-ellipsoidal clusters and does not require the number of clusters to be pre-specified. The mean shift clusters are determined by following the initial observations along the density gradient ascent paths to the cluster centre. For details of the computation and the bandwidth selection procedure of the kernel mean shift clustering, see `?ks::kms`. The bandwidth matrix of smoothing parameters is computed as in `ks::kdde(deriv_order=1)`. ## Examples ```r ## tidy 2-d mean shift clustering library(ggplot2) data(crabs, package="MASS") crabs2 <- dplyr::select(crabs, FL, CW) t1 <- tidy_kms(crabs2) ## convex hulls of clusters t2 <- dplyr::group_by(t1, label) t2 <- dplyr::slice(t2, chull(FL,CW)) gt <- ggplot(t1, aes(x=FL, y=CW)) gt + geom_point(aes(colour=label)) + geom_polygon(data=t2, aes(fill=label), alpha=0.1, col=1, linetype="dotted") ## geospatial mean shift clustering data(wa) data(grevilleasf) hakeoides <- dplyr::filter(grevilleasf, species=="hakeoides") s1 <- st_kms(hakeoides) ## convex hulls of clusters s2 <- dplyr::group_by(s1$sf, label) s2 <- dplyr::summarise(s2, geometry=sf::st_combine(geometry)) s2 <- sf::st_convex_hull(s2) ## base R plot xlim <- c(1.2e5, 1.1e6); ylim <- c(6.1e6, 7.2e6) plot(wa, xlim=xlim, ylim=ylim) plot(s1, add=TRUE, pch=16) plot(s2, add=TRUE, lty=3, pal=function(.){ colorspace::qualitative_hcl(n=., palette="Set2", alpha=0.15)}) ## geom_sf plot gs <- ggplot(s1) + geom_sf(data=wa, fill=NA) + ggthemes::theme_map() gs + geom_sf(data=s1$sf, aes(colour=label), alpha=0.5) + geom_sf(data=s2, aes(fill=label), linetype="dotted", alpha=0.15) + coord_sf(xlim=xlim, ylim=ylim) ```

tidyst_kms function