partialpro.varpro.RdObtain the partial effect of x-variables from a VarPro analysis.
partialpro(object, xvar.names, nvar,
target, learner, newdata, method = c("unsupv", "rnd", "auto"),
verbose = FALSE, papply = mclapply, ...)VarPro object obtained from previous call to varpro.
Names of the x-variables to be used.
Number of variables to be used. Default is all.
For classification, an integer or character value specifying the class that the partial effect is computed for (defaults to the last class).
Optional function for passing a user's personally constructed learner (see details below).
Optional data frame containing x-data used for the analysis. If present, partial effects are obtained for these data points. Otherwise the default action is to use the training data.
Isolation forest method used for unlimited virtual
twins. Choices are "unsupv" (unsupervised analysis, the default),
"rnd" (pure random splitting) and "auto" (auto-encoder, a type of
multivariate forest). See isopro for details.
Print verbose output?
Use mclapply or lapply.
Additional hidden options: "cut", "nsmp", "nvirtual", "nmin", "alpha", "df", "sampsize", "ntree", "nodesize", "mse.tolerance".
Obtain partial effects for requested variables from a VarPro analysis. Some variables will have been filtered in the VarPro analysis and in such cases their partial effects cannot be obtained (as filtered variables are considered to be noisy variables).
Partial effects are obtained using predicted values from the random
forest object built during the VarPro analysis. Unlimited
Virtual Twins (UVT) are used to restrict the prediction call to
partial plot data (i.e. the artificially created data used for the
partial effects estimation) that meet an isolation forest criteria
specified by the internal option cut, a small percentile
value which filters unlikely to occur partial data. Isolation
forests are constructed via the function isopro.
For regression, the partial effect is on the scale of the response; for survival it is either mortality (default) or RMST (the latter when the user specifies this in the original varpro call: see examples below); for classification it is the log-odds for the specified target class label (see examples below).
Partial effects are estimated locally using polynomial linear models
fit to the predicted values from the learner. Degrees of freedom of
the local polynomial model is set by option df (the default is
2 corresponding to a quadratic model).
By default, the random forest built during the VarPro analysis is used
for prediction, however users can provide their own predictor using
option learner. The user's learner should be trained using the
training data used in the varpro analysis. learner then takes
the form of a function whose input is a data frame containing feature
values and whose output is the predicted value from the user's
learner. For regression and survival, the output should be a vector.
For classification, the output should be matrix whose columns are the
predicted probability for each class label given in the order of the
original class outcome. If the input is missing, the function should
return the predicted values for the original training data. See the
examples below which illustrate using an external: (1) random forest
learner (which can sometimes be better than the random forests learner built
during the VarPro analysis); (2) gradient tree boosting learner; and
(3) BART (Bayesian Adaptive Regression Trees).
Ishwaran H. (2025). Multivariate Statistics: Classical Foundations and Modern Machine Learning, CRC (Chapman and Hall), in press.
##------------------------------------------------------------------
##
## Boston housing
##
##------------------------------------------------------------------
library(mlbench)
data(BostonHousing)
par(mfrow=c(2,3))
plot((oo.boston<-partialpro(varpro(medv~.,BostonHousing),nvar=6)))
# \donttest{
##------------------------------------------------------------------
##
## Boston housing using newdata option
##
##
##------------------------------------------------------------------
library(mlbench)
data(BostonHousing)
o <- varpro(medv~.,BostonHousing)
par(mfrow=c(2,3))
plot(partialpro(o,nvar=3))
## same but using newdata (set to first 6 cases of the training data)
plot(partialpro(o,newdata=o$x[1:6,],nvar=3))
##------------------------------------------------------------------
##
## Boston housing with externally constructed rf learner
##
##------------------------------------------------------------------
## varpro analysis
library(mlbench)
data(BostonHousing)
o <- varpro(medv~.,BostonHousing)
## default partial pro call
pro <- partialpro(o, nvar=3)
## partial pro call using built in rf learner
mypro <- partialpro(o, nvar=3, learner=rf.learner(o))
## compare the two
par(mfrow=c(2,3))
plot(pro)
plot(mypro, ylab="external rf learner")
##------------------------------------------------------------------
##
## Boston housing: tree gradient boosting learner, bart learner
##
##------------------------------------------------------------------
if (library("gbm", logical.return=TRUE) &&
library("BART", logical.return=TRUE)) {
## varpro analysis
library(parallel)
library(mlbench)
data(BostonHousing)
o <- varpro(medv~.,BostonHousing)
## default partial pro call
pro <- partialpro(o, nvar=3)
## partial pro call using built in gradient boosting learner
## mypro <- partialpro(o, nvar=3, learner=gbm.learner(o, n.trees=1000, n.cores=detectCores()))
## The only way to pass check-as-cran
mypro <- partialpro(o, nvar=3, learner=gbm.learner(o, n.trees=1000, n.cores=2))
## partial pro call using built in bart learner
## mypro2 <- partialpro(o, nvar=3, learner=bart.learner(o, mc.cores=detectCores()))
## The only way to pass check-as-cran
mypro2 <- partialpro(o, nvar=3, learner=bart.learner(o, mc.cores=2))
## compare the learners
par(mfrow=c(3,3))
plot(pro)
plot(mypro, ylab="external boosting learner")
plot(mypro2, ylab="external bart learner")
}
##------------------------------------------------------------------
##
## peak vo2 with 5 year rmst
##
##------------------------------------------------------------------
data(peakVO2, package = "randomForestSRC")
par(mfrow=c(2,3))
plot((oo.peak<-partialpro(varpro(Surv(ttodead,died)~.,peakVO2,rmst=5),nvar=6)))
##------------------------------------------------------------------
##
## veteran data set with celltype as a factor
##
##------------------------------------------------------------------
data(veteran, package = "randomForestSRC")
dta <- veteran
dta$celltype <- factor(dta$celltype)
par(mfrow=c(2,3))
plot((oo.veteran<-partialpro(varpro(Surv(time, status)~., dta), nvar=6)))
##------------------------------------------------------------------
##
## iris: classification analysis showing partial effects for all classes
##
##------------------------------------------------------------------
o.iris <- varpro(Species~.,iris)
yl <- paste("log-odds", levels(iris$Species))
par(mfrow=c(3,2))
plot((oo.iris.1 <- partialpro(o.iris, target=1, nvar=2)),ylab=yl[1])
plot((oo.iris.2 <- partialpro(o.iris, target=2, nvar=2)),ylab=yl[2])
plot((oo.iris.3 <- partialpro(o.iris, target=3, nvar=2)),ylab=yl[3])
##------------------------------------------------------------------
##
## iowa housing data
##
##------------------------------------------------------------------
## quickly impute the data; log transform the outcome
data(housing, package = "randomForestSRC")
housing <- randomForestSRC::impute(SalePrice~., housing, splitrule="random", nimpute=1)
dta <- data.frame(data.matrix(housing))
dta$y <- log(housing$SalePrice)
dta$SalePrice <- NULL
## partial effects analysis
o.housing <- varpro(y~., dta, nvar=Inf)
oo.housing <- partialpro(o.housing,nvar=15)
par(mfrow=c(3,5))
plot(oo.housing)
# }