sliceplot: Plot Slices of Bivariate Functions

Description

This function plots slices from user defined values of bivariate surfaces.

Usage

sliceplot(x, y = NULL, z = NULL, view = 1, c.select = NULL,
  values = NULL, probs = c(0.1, 0.5, 0.9), grid = 100,
  legend = TRUE, pos = "topright", digits = 2, data = NULL,
  rawdata = FALSE, type = "interp", linear = FALSE,
  extrap = FALSE, k = 40, rug = TRUE, rug.col = NULL,
  jitter = TRUE, ...)

Arguments

x: a matrix or data frame, containing the covariates for which the effect should be plotted in the first and second column and at least a third column containing the effect, typically the structure for bivariate functions returned within bayesx and read.bayesx.output model term objects is used, also see fitted.bayesx. Another possibility is to specify the plot via a formula, e.g. for simple plotting of bivariate surfaces z ~ x + y, also see the example.
y: if x is a vector the argument y and z must also be supplied as vectors.
z: if x is a vector the argument y and z must also be supplied as vectors, z defines the surface given by \(z = f(x, y)\).
view: which variable should be used for the x-axis of the plot, the other variable will be used to compute the slices. May also be a character with the name of the corresponding variable.
c.select: integer, selects the column that is used in the resulting matrix to be used as the z argument.
values: the values of the x or y variable that should be used for computing the slices, if set to NULL, slices will be constructed according to the quantiles, see also argument probs.
probs: numeric vector of probabilities with values in [0,1] to be used within function quantile to compute the values for plotting the slices.
grid: the grid size of the surface where the slices are generated from.
legend: if set to TRUE, a legend with the values that where used for slicing will be added.
pos: the position of the legend, see also function legend.
digits: the decimal place the legend values should be rounded.
data: if x is a formula, a data.frame or list. By default the variables are taken from environment(x): typically the environment from which plot3d is called.
rawdata: if set to TRUE, the data will not be interpolated, only raw data will be used. This is useful when displaying data on a regular grid.
type: character. Which type of interpolation metjod should be used. The default is type = "interp", see function interp. The two other options are type = "mba", which calls function mba.surf of package MBA, or type = "mgcv", which uses a spatial smoother withing package mgcv for interpolation. The last option is definitely the slowest, since a full regression model needs to be estimated.
linear: logical. Should linear interpolation be used withing function interp?
extrap: logical. Should interpolations be computed outside the observation area (i.e., extrapolated)?
k: integer. The number of basis functions to be used to compute the interpolated surface when type = "mgcv".
rug: add a rug to the plot.
jitter: if set to TRUE a jittered rug plot is added.
rug.col: specify the color of the rug representation.
...: parameters passed to matplot and legend.

Author

Nikolaus Umlauf, Thomas Kneib, Stefan Lang, Achim Zeileis.

Details

Similar to function plot3d, this function first applies bivariate interpolation on a regular grid, afterwards the slices are computed from the resulting surface.

Examples

Run this code

## generate some data
set.seed(111)
n <- 500

## regressors
dat <- data.frame(z = runif(n, -3, 3), w = runif(n, 0, 6))

## response
dat$y <- with(dat, 1.5 + cos(z) * sin(w) + rnorm(n, sd = 0.6))

if (FALSE) {
## estimate model
b <- bayesx(y ~ sx(z, w, bs = "te", knots = 5), data = dat, method = "REML")
summary(b)

## plot estimated effect
plot(b, term = "sx(z,w)", sliceplot = TRUE)
plot(b, term = "sx(z,w)", sliceplot = TRUE, view = 2)
plot(b, term = "sx(z,w)", sliceplot = TRUE, view = "w")
plot(b, term = "sx(z,w)", sliceplot = TRUE, c.select = 4)
plot(b, term = "sx(z,w)", sliceplot = TRUE, c.select = 6)
plot(b, term = "sx(z,w)", sliceplot = TRUE, probs = seq(0, 1, length = 10))
}

## another variation
dat$f1 <- with(dat, sin(z) * cos(w))
sliceplot(cbind(z = dat$z, w = dat$w, f1 = dat$f1))

## same with formula 
sliceplot(sin(z) * cos(w) ~ z + w, ylab = "f(z)", data = dat)

## compare with plot3d()
plot3d(sin(z) * 1.5 * w ~ z + w, zlab = "f(z,w)", data = dat)
sliceplot(sin(z) * 1.5 * w ~ z + w, ylab = "f(z)", data = dat)
sliceplot(sin(z) * 1.5 * w ~ z + w, view = 2, ylab = "f(z)", data = dat)