Skip to contents

Add a fitted regression curve or convex hulls to a plot

Source: R/plotcmds.R
tda_pl_regression.Rd

tda_pl_regression draws the fitted curve that TDA's screen-only xreg command used to add to an X11 scatterplot – the X11 front end was never part of this source tree, so the three fits it offered are reproduced here and drawn through the ordinary PostScript session: a lowess curve, a least-squares line, or an L1-norm (least-absolute-deviations) line, one per group. The lowess curve is TDA's, drawn by scplot – the same one tda_pl_scatter draws, and the one xreg drew. (npreg(opt = 4) runs the same C routine but cannot give the same curve: it resets its band width to 1 when it is below epsilon, before the lowess branch, so the delta shortcut is never 0, while scplot passes 0.) The least-squares line is ordinary regression of y on x; the L1 line is TDA's l1reg via tda_l1reg. Like xreg, the straight-line fits are drawn from the group's smallest x to its largest.

Usage

tda_pl_regression(
  p,
  x,
  y,
  type = c("lowess", "least_squares", "l1"),
  by = NULL,
  bandwidth = 0.5,
  lty = NULL,
  lw = NULL,
  gray = NULL,
  ...
)

tda_pl_hull(
  p,
  x,
  y,
  by = NULL,
  lty = NULL,
  lw = NULL,
  gray = NULL,
  symbol = NULL,
  size = NULL,
  smooth = NULL,
  expand = NULL,
  ...
)

Arguments

p

A plot session started with tda_ps.

x, y

Column names in the session's data.

type

The fit: "lowess" (the default, as in xreg), "least_squares", or "l1".

by

Optional name of a grouping column; one fit (or hull) per group. Without it, all points are one group.

bandwidth

The lowess span, xreg's sig= (default 0.5, clamped there to (0, 1)).

lty, lw, gray

Line type, width (mm), and gray level for the drawn curve.

...

Further options passed to the drawing command.

symbol, size

for tda_pl_hull, a marker symbol and its size in millimetres drawn at the points; plotch's s= and fs=.

smooth

the number of intervals for Akima smoothing of the hull's outline; plotch's ns=.

expand

a margin added to a smoothed hull, in millimetres; TDA's ic=. Each point of the smoothed outline is pushed away from the hull's centroid in proportion to its distance from it along each axis, so the outline grows without changing shape (t_plot.c, pl_plotch).

Value

The session, with the curve(s) added. The fitted coefficients (for the straight-line types) or curves are kept in p$fits, one entry per call and group.

Details

tda_pl_hull is xconh: the convex hull of each group's points, drawn as a closed outline.

The third screen command, xplotf, plotted values straight from a file; tda_pl_scatter and tda_pl_lines with raw vectors already do that.

See also

Other plotting: plot.tda_ple(), tda_check_ps(), tda_pl(), tda_pl_arc(), tda_pl_axis(), tda_pl_graph(), tda_pl_hist, tda_pl_panel(), tda_pl_scatter(), tda_plot_ps(), tda_ps(), tda_ps3(), tda_read_ps()

Examples

set.seed(1)
d <- data.frame(x = round(runif(40, 0, 10), 2))
d$y <- round(1 + 0.8 * d$x + rnorm(40), 2)
p <- tda_ps(d, xlim = c(0, 10), ylim = c(0, 12))
p <- tda_pl_frame(p)
p <- tda_pl_scatter(p, "x", "y", symbol = 1)
p <- tda_pl_regression(p, "x", "y", type = "least_squares",
                       lty = "dashed")
p$fits
#> $least_squares
#>     alpha      beta 
#> 1.3000324 0.7544393 
#>