itree: Recursive Partitioning and Regression Trees a la rpart, but with some extensions targeted at growing interpretable/parsimonious trees.

Description

Fit a itree model.

Usage

itree(formula, data, weights, subset, na.action = na.itree,
	 method, penalty= "none", model = FALSE, 
	 x = FALSE, y = TRUE, parms, control, cost, ...)

Arguments

formula

a formula, with a response but no interaction terms.

data

an optional data frame in which to interpret the variables named in the formula.

weights

optional case weights.

subset

optional expression saying that only a subset of the rows of the data should be used in the fit.

na.action

the default action deletes all observations for which y is missing, but keeps those in which one or more predictors are missing.

method

one of "anova", "class", "extremes", "purity", "class_extremes", "class_purity", "regression_extremes", or "regression_purity". The purity and extremes methods are new to itree. Unlike rpart, itree does not currently support method="poisson" or method="exp". If method is missing then the routine tries to make an intelligent guess -- the default is the CART methodology, as in rpart. If y is a factor then method="class" is assumed, otherwise method="anova" is assumed. Passing a factor with method="purity" is equivalent to passing method="class_purity", and similarly for extremes/regression. It is wisest to specify the method directly, especially as more criteria may added to the function in future. As in rpart, method can be a list of functions named init, split and eval. See the rpart documentation for how this works.

penalty

one of "none", "newvar" or "ema". The penalty for splitting a particular node on a specified predictor given the predictors already used in the branch leading to this node. Default is "none" which corresponds to CART. "newvar" penalizes predictors not used in the branch leading to the current node. "ema" implements an exponential moving average style penalty whereby recently used variables are favored.

model

if logical: keep a copy of the model frame in the result? If the input value for model is a model frame (likely from an earlier call to the itree function), then this frame is used rather than constructing new data.

keep a copy of the x matrix in the result.

keep a copy of the dependent variable in the result. If missing and model is supplied this defaults to FALSE.

parms

optional parameters for the splitting function. Anova splitting has no parameters. For classification splitting, the list can contain any of: the vector of prior probabilities (component prior), the loss matrix (component loss) or the splitting index (component split). The priors must be positive and sum to 1. The loss matrix must have zeros on the diagonal and positive off-diagonal elements. The splitting index can be gini or information. The default priors are proportional to the data counts, the losses default to 1, and the split defaults to gini. For the regression extremes method, parms=1 or parms=-1 specifies whether we are looking for high or low means respectively (see Buja & Lee). Default is 1 for high means. For classification extremes, parms is a list specificying the class of interest -- see the examples for syntax.

control

a list of options that control details of the itree algorithm, similar to rpart.control. See itree.control.

cost

a vector of non-negative costs, one for each variable in the model. Defaults to one for all variables. These are scalings to be applied when considering splits, so the improvement on splitting on a variable is divided by its cost in deciding which split to choose. Note that costs are not currently supported by the extremes or purity methods.

...

arguments to itree.control may also be specified in the call to itree. They are checked against the list of valid arguments.

Value

An object of class itree. See itree.object.

Details

itree is based on the code of rpart, but with some extensions targeted at growing interpretable/parsimonious trees. Bug reports and the like should be directed to this package's maintainer -- not rpart's.

References

Breiman, Friedman, Olshen, and Stone. (1984) Classification and Regression Trees. Buja, Andreas and Lee, Yung-Seop (2001). Data Mining Criteria for Tree-Based Regression and Classification, Proceedings of KDD 2001, 27-36.

Wadsworth.

Examples

Run this code

#CART (same as rpart):
fit <- itree(Kyphosis ~ Age + Number + Start, data=kyphosis)
fit2 <- itree(Kyphosis ~ Age + Number + Start, data=kyphosis,
              parms=list(prior=c(.65,.35), split='information'))
fit3 <- itree(Kyphosis ~ Age + Number + Start, data=kyphosis,
              control=itree.control(cp=.05))
par(mfrow=c(1,2), xpd=NA) # otherwise on some devices the text is clipped
plot(fit)
text(fit, use.n=TRUE)
plot(fit2)
text(fit2, use.n=TRUE)

#### new to itree:
#same example, but using one-sided extremes:
fit.ext <- itree(Kyphosis ~ Age + Number + Start, data=kyphosis,method="extremes",
				parms=list(classOfInterest="absent"))
#we see buckets with every y="absent":
plot(fit.ext); text(fit.ext,use.n=TRUE) 


library(mlbench); data(BostonHousing)

#one sided purity:
fit4 <- itree(medv~.,BostonHousing,method="purity",minbucket=25)

#low means tree:
fit5 <- itree(medv~.,BostonHousing,method="extremes",parms=-1,minbucket=25)

#new variable penalty:
fit6 <- itree(medv~.,BostonHousing,penalty="newvar",interp_param1=.2)

#ema penalty
fit7 <- itree(medv~.,BostonHousing,penalty="ema",interp_param1=.1)

#one-sided-purity + new variable penalty:
fit8 <- itree(medv~.,BostonHousing,method="purity",penalty="newvar",interp_param1=.2)

#one-sided extremes for classification must specify a "class of interest"
data(PimaIndiansDiabetes)
levels(PimaIndiansDiabetes$diabetes)  
fit9.a <- itree(diabetes~.,PimaIndiansDiabetes,minbucket=50,
                 method="extremes",parms=list(classOfInterest="neg"))
                 
plot(fit9.a); text(fit9.a)

#can also pass the index of the class of interest in levels().
fit9.b <- itree(diabetes~.,PimaIndiansDiabetes,minbucket=50,
                 method="extremes",parms=list(classOfInterest=1))
# so fit9.a = fit9.b

Run the code above in your browser using DataLab

Get 50% off unlimited learning