plotpc: Plot principal component histograms around a scatter plot

Description

Plot principal component histograms around the scatter plot of two variables. Mostly useful as a tool for teaching principal components.

Usage

plotpc(x, xrange=NULL, hist=TRUE, main="Principal components", xlab=NULL, ylab=NULL, gp.points=gpar(cex=.6), pch=20, height=xrange/10, breaks="Sturges", adjust=1, gp.hist=if(hist) gp.hist <- gpar(col="gray", fill="gray") else     gp.hist <- gpar(col="black"), gp.text=gpar(cex=.8, font=2), gp.axis=gpar(col="gray", lwd=2), sd.ellipse=NA, gp.ellipse=gpar(col="gray", lwd=2), heightx=NULL, breaksx=NULL, adjustx=NULL, gp.histx=NULL, textx="", gp.textx=NULL, axis.lenx=0, gp.axisx=NULL, heighty=NULL, breaksy=NULL, adjusty=NULL, gp.histy=NULL, texty="", gp.texty=NULL, axis.leny=0, gp.axisy=NULL, height1=NULL, flip1=FALSE, breaks1=NULL, adjust1=NULL, gp.hist1=NULL, offset1=NULL, text1=NULL, gp.text1=NULL, axis.len1=2, gp.axis1=NULL, height2=NULL, flip2=FALSE, breaks2=NULL, adjust2=NULL, gp.hist2=NULL, offset2=NULL, text2=NULL, gp.text2=NULL, axis.len2=2, gp.axis2=NULL, angle3=NA, height3=NULL, flip3=FALSE, breaks3=NULL, adjust3=NULL, gp.hist3=NULL, offset3=NULL, text3=NULL, gp.text3=NULL, axis.len3=0, gp.axis3=NULL, angle4=NA, height4=NULL, flip4=FALSE, breaks4=NULL, adjust4=NULL, gp.hist4=NULL, offset4=NULL, text4=NULL, gp.text4=NULL, axis.len4=0, gp.axis4=NULL, angle5=NA, height5=NULL, flip5=FALSE, breaks5=NULL, adjust5=NULL, gp.hist5=NULL, offset5=NULL, text5=NULL, gp.text5=NULL, axis.len5=0, gp.axis5=NULL, angle6=NA, height6=NULL, flip6=FALSE, breaks6=NULL, adjust6=NULL, gp.hist6=NULL, offset6=NULL, text6=NULL, gp.text6=NULL, axis.len6=0, gp.axis6=NULL, angle7=NA, height7=NULL, flip7=FALSE, breaks7=NULL, adjust7=NULL, gp.hist7=NULL, offset7=NULL, text7=NULL, gp.text7=NULL, axis.len7=0, gp.axis7=NULL, yonx = FALSE, offset.yonx=-xrange/2.5, text.yonx="y~x", gp.text.yonx=NULL, axis.len.yonx=xrange/2.5, gp.axis.yonx=gpar(col=1), xony = FALSE, offset.xony=-xrange/2.5, text.xony="x~y", gp.text.xony=NULL, axis.len.xony=xrange/2.5, gp.axis.xony=gpar(col=1))

Arguments

A two column matrix or dataframe. The principal components of the x will be calculated treating each column as a variable.

hist

Default TRUE to plot histograms. Set to FALSE to plot densities instead. The various "histogram" arguments will then apply to densities rather than to histograms.

xrange

The range of the x axis. That is, xlim will be c(mean(x[,1]) - xrange/2, mean(x[,1]) + xrange/2), and ylim will have the same range about mean(x[,2]). Default NULL, meaning automatically deduce axis limits from the x argument.

main

Main title. Default "Principal components".

xlab

x axis label. Default NULL, meaning create the label automatically from the column names of x.

ylab

y axis label. Default NULL, meaning create the label automatically from the column names of x.

gp.points

Graphic parameters for the plotted points. Default gpar(cex=.6).

pch

Plot character for the plotted points. Default 20. The following arguments apply to all histograms. These can be overridden by using the histogram-specific argument e.g. override the height argument for the first principal component by specifying height1.

height

Height of histograms. Default xrange/10. Use a negative height to flip a histogram around its base.

breaks

Passed on to hist. Default "Sturges". Using something like breaks=12 can be useful.

adjust

Passed on to density. Default 1. Use something like adjust=.5 for more details in the density plots.

gp.hist

Graphic parameters for the histograms or densities. If hist==TRUE then the default is gpar(col="gray", fill="gray") where col is the color of the lines delineating the histograms, and fill is the color filling the histograms. If hist==FALSE then the default is gpar(col="black").

gp.axis

Graphic parameters for the axis drawn through the scatter of points. Default gpar(col="gray", lwd=2) meaning draw the axes as thickish gray lines.

sd.ellipse

If greater than 0, draw a confidence ellipse for the principal components at sd.ellipse standard deviations. Default is NA, meaning do not draw an ellipse.

gp.ellipse

Graphic parameters for the ellipse. Default gpar(col="gray", lwd=2).

gp.text

Graphic parameters for text above the histograms. Default gpar(cex=.8, font=2). The following arguments apply to the histogram on the x axis.

heightx

Default NULL, meaning use height. Use 0 to not plot the x histogram.

breaksx

Default NULL, meaning use breaks.

adjustx

Default NULL, meaning use adjust.

gp.histx

Default NULL, meaning use gp.hist.

textx

Text drawn above the histogram. Default "", meaning no text. The text is drawn using gp.textx.

gp.textx

Graphic parameters for the text above the histogram. Default NULL, meaning use gp.text.

axis.lenx

Length of horizontal line drawn through the center of the points. Units are standard deviations of x[,1]. Default 0, meaning do not plot a horizontal axis.

gp.axisx

Default NULL, meaning use gp.axis.

heighty, breaksy, adjusty, gp.histy, texty, gp.texty, axis.leny, gp.axisy

As above but for the histogram on the y axis. The following arguments apply to the first principal component.

height1

Default NULL, meaning use height. Use 0 to not plot the histogram for the first principal component.

flip1

Flip the position of the histogram around the axis of the first principal component. Default FALSE, meaning do not flip.

breaks1

Default NULL, meaning use breaks.

adjust1

Default NULL, meaning use adjust.

gp.hist1

Default NULL, meaning use gp.hist.

offset1

Distance of the histogram plot from the center of the graph, in native units. Default NULL, meaning automatic.

text1

Text drawn above the histogram. Default NULL, meaning generate the text automatically. Use "" for no text. The text is drawn using gp.text1.

gp.text1

Graphic parameters for the text above the histogram. Default NULL, meaning use gp.text.

axis.len1

Length of line drawn along the first principal axis. Units are standard deviations of the points projected onto that axis. Default 2, meaning draw a line of length plus and minus two standard deviations. Use 0 for no axis.

gp.axis1

Default NULL, meaning use gp.axis.

height2, flip2, breaks2, adjust2, gp.hist2, offset2, text2, gp.text2, axis.len2, gp.axis2

As above but for the second principal component. The following arguments apply to the optional histogram at angle3. By default, angle3=NA, meaning do not plot the histogram. Use, say, angle3=45 to plot a histogram at 45 degrees. By setting angle3 to angle7 you can plot up to five extra histograms at any angles.

angle3

Default NA, meaning do not plot a histogram. Use, say, angle3=45 to plot a histogram at 45 degrees.

height3

Default NULL, meaning use height.

flip3

Default FALSE.

breaks3

Default NULL, meaning use breaks.

adjust3

Default NULL, meaning use adjust.

gp.hist3

Default NULL, meaning use gp.hist.

offset3

Default NULL, meaning automatic.

text3

Default NULL, meaning automatic.

gp.text3

Default NULL, meaning use gp.text.

axis.len3

Length of axis drawn at angle3 through the scatter of points. Default 0, meaning do not plot the axis.

gp.axis3

Default NULL, meaning use gp.axis.

angle4, height4, flip4, breaks4, adjust4, gp.hist4, offset4, text4, gp.text4, axis.len4, gp.axis4

As above but for the angle4 histogram.

angle5, height5, flip5, breaks5, adjust5, gp.hist5, offset5, text5, gp.text5, axis.len5, gp.axis5

As above but for the angle5 histogram.

angle6, height6, flip6, breaks6, adjust6, gp.hist6, offset6, text6, gp.text6, axis.len6, gp.axis6

As above but for the angle6 histogram.

angle7, height7, flip7, breaks7, adjust7, gp.hist7, offset7, text7, gp.text7, axis.len7, gp.axis7

As above but for the angle7 histogram. The following arguments apply to the optional "y on x" regression line.

yonx

TRUE to plot a "y on x" linear regression line. Default FALSE.

offset.yonx

Position of text plotted on regression line. Default -xrange/2.5.

text.yonx

Text plotted on the regression line. Default "y~x".

gp.text.yonx

Graphic parameters for the text plotted on the regression line. Default NULL, meaning use gp.text.

axis.len.yonx

Length of regression line in gpar "native" units. Default -xrange/2.5.

gp.axis.yonx

Graphic parameters for the regression line. Default gpar(col=1).

xony, offset.xony, text.xony, gp.text.xony, axis.len.xony, gp.axis.xony

As above but for a "x on y" regression.

Value

Invisibly returns the viewport used to create the plotpc axes. This allows you to add text using the "native" coordinates of the plot. See the examples below.

Examples

Run this code

data(iris)
x <- iris[,c(3,4)] # select Petal.Length and Petal.Width
plotpc(x, main="Example 1\n")

# example with some parameters and showing densities
plotpc(x,
       main="Example 2:\nPrincipal component densities\n",
       hist=FALSE,                    # plot densities not histograms
       adjust=.5,                     # finer resolution in the density plots
       gp.axis=gpar(lty=3),           # gpar of axes
       heightx=0,                     # don't display x histogram
       heighty=0,                     # don't display y histogram
       text1="Principal Component 1", # text above hist for 1st principal component
       text2="Principal Component 2", # text above hist for 2nd principal component
       axis.len2=4,                   # length of 2nd principal axis (in std devs)
       offset1=2.5,                   # offset of component 1 density plot
       offset2=5)                     # offset of component 2 density plot

# example using "angles"
vp <- plotpc(x,
       main="Example 3:\nProjections\n",
       xrange=25,      # give ourselves some space
       heightx=0,      # don't display x histogram
       heighty=0,      # don't display y histogram
       angle3=-60,     # project at -60 degrees
       angle4=-25,     # project at -25 degrees
       angle5=20,      # project at 20 degrees
       angle6=70)      # project at 70 degrees

# add text to the graph, can use native coords
pushViewport(vp)
grid.text("Projections at\nvarious angles",
          x=unit(10, "native"), y=unit(12.5, "native"),
          gp=gpar(col="red"))
popViewport()

# example showing principal axes
x <- iris[iris$Species=="versicolor",c(3,4)]
vp <- plotpc(x,
       main="Example 4:\nPrincipal axes with confidence ellipse\n",
       sd.ellipse=2,                       # ellipse at two standard devs
       heightx=0, heighty=0, height1=0, height2=0, # no histograms
       gp.ellipse=gpar(col=1),             # ellipse in black
       axis.lenx=4, axis.leny=5,           # lengthen horiz and vertical axes
       axis.len1=4, gp.axis1=gpar(col=1),  # lengthen pc1 axis, draw in black
       axis.len2=8, gp.axis2=gpar(col=1))  # lengthen pc2 axis, draw in black

pushViewport(vp) # add text to the graph
un <- function(x) unit(x, "native")
grid.text("PC1", x=un(2.2), y=un(.6),   gp=gpar(cex=.8, font=2))
grid.text("PC2", x=un(3.9), y=un(2.35), gp=gpar(cex=.8, font=2))
grid.text("X1",  x=un(2.2), y=un(1.4),  gp=gpar(cex=.8, font=2))
grid.text("X2",  x=un(4.3), y=un(2.5),  gp=gpar(cex=.8, font=2))
popViewport()

# example comparing linear regression to principal axis
x <- iris[iris$Species=="setosa",c(3,4)]
vp <- plotpc(x,
       main="Example 5:\nRegression lines and\nfirst principal component",
       heightx=0, heighty=0, height1=0, height2=0, # no histograms
       gp.points=gpar(col="steelblue"),      # color of points
       axis.len1=4,  gp.axis1=gpar(col="gray", lwd=3),
       axis.len2=.15, gp.axis2=gpar(col=1),  # just a little blip of an axis
       yonx=TRUE, xony=TRUE)                 # display regression lines

pushViewport(vp) # add text to the principal component line
grid.text("PC1", x=unit(.8, "native"), y=unit(0, "native"),
          gp=gpar(col="gray", cex=.8, font=2))
popViewport()

Run the code above in your browser using DataCamp Workspace

Description

Usage

Arguments

Value

See Also

Examples