vis.scam: Visualization of SCAM objects

Description

Produces perspective or contour plot views of scam model predictions. The code is a clone of vis.gam of the mgcv package.

Usage

vis.scam(x,view=NULL,cond=list(),n.grid=30,too.far=0,col=NA,
        color="heat",contour.col=NULL,se=-1,type="link",
        plot.type="persp",zlim=NULL,nCol=50,...)

Value

Simply produces a plot.

Arguments

The documentation below is the same as in documentation object vis.gam.

x: a scam object, produced by scam()
view: an array containing the names of the two main effect terms to be displayed on the x and y dimensions of the plot. If omitted the first two suitable terms will be used.
cond: a named list of the values to use for the other predictor terms (not in view). Variables omitted from this list will have the closest observed value to the median for continuous variables, or the most commonly occuring level for factors. Parametric matrix variables have all the entries in each column set to the observed column entry closest to the column median.
n.grid: The number of grid nodes in each direction used for calculating the plotted surface.
too.far: plot grid nodes that are too far from the points defined by the variables given in view can be excluded from the plot. too.far determines what is too far. The grid is scaled into the unit square along with the view variables and then grid nodes more than too.far from the predictor variables are excluded.
col: The colours for the facets of the plot. If this is NA then if se>0 the facets are transparent, otherwise the colour scheme specified in color is used. If col is not NA then it is used as the facet colour.
color: the colour scheme to use for plots when se<=0. One of "topo", "heat", "cm", "terrain", "gray" or "bw". Schemes "gray" and "bw" also modify the colors used when se>0.
contour.col: sets the colour of contours when using plot.type="contour". Default scheme used if NULL.
se: if less than or equal to zero then only the predicted surface is plotted, but if greater than zero, then 3 surfaces are plotted, one at the predicted values minus se standard errors, one at the predicted values and one at the predicted values plus se standard errors.
type: "link" to plot on linear predictor scale and "response" to plot on the response scale.
plot.type: one of "contour" or "persp".
zlim: a two item array giving the lower and upper limits for the z-axis scale. NULL to choose automatically.
nCol: The number of colors to use in color schemes.
...: other options to pass on to persp, image or contour.

Author

Simon Wood simon.wood@r-project.org

Examples

Run this code

library(scam)

# Example with factor variable
set.seed(0)
fac<-rep(1:4,20)
x <- runif(80)*5;
y <- fac+log(x)/5+rnorm(80)*0.1
fac <- factor(fac)
b <- scam(y~fac+s(x,bs="mpi"))

vis.scam(b,theta=-35,color="heat") # factor example

# Example with "by" variable

z<-rnorm(80)*0.4   
y<-as.numeric(fac)+log(x)*z+rnorm(80)*0.1
b<-scam(y~fac+s(x,by=z))
g <- gam(y~fac+s(x,by=z))

vis.scam(b,theta=-35,color="terrain",cond=list(z=1)) # by variable example
vis.scam(b,view=c("z","x"),theta= 65) # plot against by variable
## compare with gam(mgcv)...
vis.gam(g,theta=-35,color="terrain",cond=list(z=1)) # by variable example
vis.gam(g,view=c("z","x"),theta= 65) # plot against by variable

## all three smooths are increasing...
set.seed(2)
n <- 400
x <- runif(n, 0, 1)
f1 <- log(x *5)
f2 <-  exp(2 * x) - 4
f3 <-  5* sin(x)
e <- rnorm(n, 0, 2)
fac <- as.factor(sample(1:3,n,replace=TRUE))
fac.1 <- as.numeric(fac==1)
fac.2 <- as.numeric(fac==2)
fac.3 <- as.numeric(fac==3)
y <- f1*fac.1 + f2*fac.2 + f3*fac.3 + e 
dat <- data.frame(y=y,x=x,fac=fac,f1=f1,f2=f2,f3=f3)

b1 <- scam(y ~ s(x,by=fac,bs="mpi"),data=dat,optimizer="efs")
plot(b1,pages=1,scale=0,shade=TRUE)
summary(b1)
vis.scam(b1,theta=-40,color="terrain",cond=list(z=1))

## note that the preceding, b1, fit is the same as....
b2 <- scam(y ~ s(x,by=as.numeric(fac==1),bs="mpi")+s(x,by=as.numeric(fac==2),bs="mpi")+
    s(x,by=as.numeric(fac==3),bs="mpi"),data=dat,optimizer="efs")
summary(b2)

## Note that as in gam() when using factor 'by' variables, centering
## constraints are applied to the smooths, which usually means that the 'by'
## variable should be included as a parametric term, as well. 
## The difference with scam() is that here a 'zero intercept' constraint is 
## applied in place of 'centering' (although scam's fitted smooths are centred for plotting).
## compare with the gam() fits..  
g1 <- gam(y ~ fac+s(x,by=fac),data=dat)
g2 <- gam(y ~ s(x,by=fac),data=dat)
summary(g1)
summary(g2)
plot(g1,pages=1,scale=0,shade=TRUE)