burl.tree(tree, nodes, screen.arg=<<see below>>,
          figs=c(1, length(xlevels)), plot = T)

tree:

fitted model object of class "tree". This is assumed to be the result of a function that produces an object with the same components as those returned by the function tree.

nodes:

an integer representing the index (node number) of the node to be examined. If missing, user selects nodes interactively as described below.

screen.arg:

integer denoting the screen to be used for the plotting. By default the active current screen plus 1 (the next one) is used. See split.screen and tree.screens for more details.

figs:

the arrangement of sub-screens to further split the screen for the plotting of the individual variable plots. Generally, a vector of the form c(n,m) where n and m are integers. By default, all sub-plots are placed in a row.

plot:

logical flag; if nodes is given, should plotting be done?

Each component of this list contains details of the competition for best split at that node in a data frame whose rows represent each potential split and with the following columns:

x:

this column is named after the predictor analyzed and contains the cutpoints or sequence numbers (subset splits).

dev:

vector of deviance change if node is split at x.

numl:

the number of observations in the left-hand split at each x.

cutleft:

in the case of a factor predictor, character vector of left-hand splits. These factor predictors are encoded as explained below in the DETAILS section.

cutright:

for factor predictors only, a character vector of right-hand splits.

Warnings result from selecting leaf nodes. Clicking the exit button will stop the burling process and return the list described above for the last node selected. See .Device and split.screen for specific details on graphic input and split-screen mode.

The strings of letters labeling splits for factor predictors are determined by a one-to-one correspondence of the letters of the alphabet (contained in the S-PLUS dataset letters) and the factor levels which can be obtained using the function levels. For example, "bd" represents the second and fourth levels of the corresponding predictor of class "factor".

z <- tree(Mileage~Weight + Type, car.test.frame)
# open graphics device then execute the following:
tree.screens()
plot(z)
burl.tree(z)

z.cu <- tree(cu.summary, na.action = na.omit)
# burl.tree() can be used non-interactively
b3 <- burl.tree(z.cu, 3) # compute alternative splits at node 3

# burl.tree provides convenient input to edit.tree
# edit according to the 6th split on Country
z.edit <- edit.tree(z.cu, b3$Country[6, ])