directlabels - utility.function - Positioning Method - far.from.others.borders

Find the point on each curve which maximizes the distance to the plot border or to another curve.

far.from.others.borders <- function(all.groups,...,debug=FALSE){ <- split(all.groups, all.groups$group)
  group.list <- list()
  for(groups in names({
    ## Run linear interpolation to get a set of points on which we
    ## could place the label (this is useful for e.g. the lasso path
    ## where there are only a few points plotted).
    approx.list <- with([[groups]], approx(x, y))
      with(approx.list, grid.points(x, y, default.units="cm"))
    group.list[[groups]] <- data.frame(approx.list, groups)
  output <- data.frame()
  for(group.i in seq_along(group.list)){ <- group.list[[group.i]]
    ## From Mark Schmidt: "For the location of the boxes, I found the
    ## data point on the line that has the maximum distance (in the
    ## image coordinates) to the nearest data point on another line or
    ## to the image boundary."
    dist.mat <- matrix(NA, length($x), 3)
    colnames(dist.mat) <- c("x","y","other")
    ## dist.mat has 3 columns: the first two are the shortest distance
    ## to the nearest x and y border, and the third is the shortest
    ## distance to another data point.
    for(xy in c("x", "y")){
      xy.vec <-[,xy]
      xy.mat <- rbind(xy.vec, xy.vec) <- get(sprintf("%slimits", xy))
      diff.mat <- xy.mat -
      dist.mat[,xy] <- apply(abs(diff.mat), 2, min)
    other.groups <- group.list[-group.i]
    other.df <-, other.groups)
    for(row.i in 1:nrow(dist.mat)){
      r <-[row.i,]
      other.dist <- with(other.df, (x-r$x)^2 + (y-r$y)^2)
      dist.mat[row.i,"other"] <- sqrt(min(other.dist))
    shortest.dist <- apply(dist.mat, 1, min)
    picked <- calc.boxes([which.max(shortest.dist),])
    ## Mark's label rotation: "For the angle, I computed the slope
    ## between neighboring data points (which isn't ideal for noisy
    ## data, it should probably be based on a smoothed estimate)."
    left <- max(picked$left, min($x))
    right <- min(picked$right, max($x))
    neighbors <- approx($x,$y, c(left, right))
    slope <- with(neighbors, (y[2]-y[1])/(x[2]-x[1]))
    picked$rot <- 180*atan(slope)/pi
    output <- rbind(output, picked)
Please contact Toby Dylan Hocking if you are using directlabels or have ideas to contribute, thanks!
Documentation website generated from source code version 2014.1.27 (svn revision 675) using inlinedocs.