ranktree {ConsRankClass}

Recursive partitioning method for the prediction of preference rankings based upon Kemeny distances

Description

Recursive partitioning method for the prediction of preference rankings based upon Kemeny distances.

Usage

ranktree(Y, X, prunplot = FALSE, control = ranktreecontrol(...), ...)

Arguments

Y

A n by m data matrix, in which there are n judges and m objects to be judged. Each row is a ranking of the objects which are represented by the columns.

X

A dataframe containing the predictor, that must have n rows.

prunplot

prunplot=TRUE returns the plot of the pruning sequence. Default value: FALSE

control

a list of options that control details of the ranktree algorithm governed by the function ranktreecontrol. The options govern the minimum size within node to split (the default value is 0.1*n, where n is the total sample size), the bound on the decrease in impurity, (default, 0.01), the algorithm chosen to compute the median ranking (default, "quick"), and other options related to the consrank algorithm, which is called by ranktree

...

arguments passed bypassing ranktreecontrol

Details

The user can use any algorithm implemented in the consrank function from the ConsRank package. All algorithms allow the user to set the option 'full=TRUE' if the median ranking(s) must be searched in the restricted space of permutations instead of in the unconstrained universe of rankings of n items including all possible ties. The output consists in a object of the class "ranktree". It contains:

X			the predictors: it must be a dataframe
Y			the response variable: the matrix of the rankings
node			a list containing teh tree-based structure:
	number		node number
	terminal		logical: TRUE is terminal node
	father		father node number of the current node
	idfather		id of the father node of the current node
	size		sample size within node
	impur		impurity at node
	wimpur		weighted impurity at node
	idatnode		id of the observations within node
	class		median ranking within node in terms of orderings
	nclass		median ranking within node in terms of rankings
	mclass		eventual multiple median rankings
	tau		Tau_x rank correlation coefficient at node
	wtau		weighted Tau_x rank correlation coefficient at node
	error		error at node
	werror		weighted error at node
	varsplit		variables generating split
	varsplitid		id of variables generating split
	cutspli		splitting point
	children		children nodes generated by current node
	idchildren		id of children nodes generated by current node
	...		other info about node
control			parameters used to build the tree
numnodes			number of nodes of the tree
tsynt			list containing the synthesis of the tree:
	children		list containing all information about leaves
	parents		list containing all information about parent nodes
geneaoly			data frame containing information about all nodes
idgenealogy			data frame containing information about all nodes in terms of nodes id
idparents			id of the parents of all the nodes
goodness			goodness -and badness- of fit measures of the tree: Tau_X, error, impurity
nomin			information about nature of the predictors
alpha			alpha parameter for pruning sequence
pruneinfo			list containing information about the pruning sequence:
	prunelist		information about the pruning
	tau		tau_X rank correlation coefficient of each subtree
	error		error of each subtree
	termnodes		number of terminal nodes of each subtree
subtrees			list of each subtree created with the cost-complexity pruning procedure

Value

An object of the class ranktree. See details for detailed information.

Author(s)

Antonio D'Ambrosio [e-mail]

References

D'Ambrosio, A., and Heiser W.J. (2016). A recursive partitioning method for the prediction of preference rankings based upon Kemeny distances. Psychometrika, vol. 81 (3), pp.774-94.

See Also

ranktreecontrol, plot.ranktree, summary.ranktree, getsubtree, validatetree, treepaths, nodepath

Examples

data("Univranks")
tree <- ranktree(Univranks$rankings,Univranks$predictors,num=50)


data(Irish)
#build the tree with default options
tree <- ranktree(Irish$rankings,Irish$predictors)

#plot the tree
plot(tree,dispclass=TRUE)

#visualize information
summary(tree)

#get information about the paths leading to terminal nodes (all the paths)
infopaths <- treepaths(tree)

#the terminal nodes
infopaths$leaves

#sample size within each terminal node
infopaths$size

#visualize the path of the second leave (terminal node number 8)
infopaths$paths[[2]]

#alternatively
nodepath(termnode=8,tree)


set.seed(132) #for reproducibility
#validation of the tree via v-fold cross-validation (default value of V=5)
vtree <- validatetree(tree,method="cv")

#extract the "best" tree
dtree <- getsubtree(tree,vtree$best_tau)

summary(dtree)

#plot the validated tree
plot(dtree,dispclass=TRUE)

#predicted rankings
rankfit <- predict(dtree,newx=Irish$predictors)

#fit of rankings
rankfit$rankings

#fit in terms of orderings
rankfit$orderings

#all info about the fit (id og the leaf, predictor values, and fit)
rankfit$orderings