find.interaction {randomSurvivalForest} | R Documentation |
Test for pairwise interactions between variables by comparing pairwise importance values to additive individual importance values.
find.interaction(x, sort = TRUE, predictorNames = NULL, n.pred = NULL, n.rep = 1, ...)
x |
An object of class (rsf, grow) . |
sort |
Should variables be sorted by importance values (only
applies if importance values are available and
predictorNames =NULL)? Default is TRUE. |
predictorNames |
Character vector of variable names. Only these variables will be considered. Default is all. |
n.pred |
Number of variables to be plotted (only applies when
predictorNames =NULL). Default is all. |
n.rep |
Number of Monte Carlo replicates. |
... |
Further arguments passed to or from other methods. |
Pairwise interactions between the first n.pred
variables
are investigated (if sort
=TRUE analysis is based on the
first n.pred
as ordered by importance values). Variables
are paired and then removed from the model. The increase in error
rate is computed and is defined to be the 'Paired' importance value
of the two variables. The increase in error rate if each
variable is removed individually is also computed and the sum of
these two values yields an 'Additive' importance value. Large
positive or negative differences between 'Paired' and 'Additive'
values indicate potentially interesting associations.
Computations can be fairly heavy if the data is large, thus users
should consider setting n.pred
to a fairly small value to
gauge computational times.
If n.rep
is greater than 1, the analysis is replicated
n.rep
times. Results reported are averaged values in this
case.
Invisibly, the interaction table.
Hemant Ishwaran hemant.ishwaran@gmail.com and Udaya B. Kogalur ubk2101@columbia.edu
H. Ishwaran (2007). Variable importance in binary trees. Cleveland Clinic Technical Report.
rsf
,
predict.rsf
.
data(veteran, package = "randomSurvivalForest") v.out <- rsf(Survrsf(time,status)~., veteran, ntree = 1000) find.interaction(v.out, n.pred = 2, n.rep=1)