predefinedClassifiers: TunePareto wrappers for certain classifiers

Description

Creates TunePareto classifier objects for the k-Nearest Neighbour classifier, support vector machines, and trees.

Usage

tunePareto.knn()
               
tunePareto.svm()
               
tunePareto.tree()
                
tunePareto.randomForest()
tunePareto.NaiveBayes()

Arguments

Value

Returns objects of class TuneParetoClassifier as described in tuneParetoClassifier. These can be passed to functions like tunePareto or trainTuneParetoClassifier.

Details

tunePareto.knn encapsulates a k-Nearest Neighbour classifier as defined in link[class]{knn} in package class. The classifier allows for supplying and tuning the following parameters of link[class]{knn}:

k, l, use.all

tunePareto.svm encapsulates the support vector machine svm classifier in package e1071. The classifier allows for supplying and tuning the following parameters:

kernel, degree, gamma, coef0, cost, nu, class.weights, cachesize, tolerance, epsilon, scale, shrinking, fitted, subset, na.action

tunePareto.tree encapsulates the CART classifier tree in package tree. The classifier allows for supplying and tuning the following parameters:

weights, subset, na.action, method, split, mincut, minsize, mindev

as well as the type parameter of predict.tree.

tunePareto.randomForest encapsulates the randomForest classifier in package randomForest. The classifier allows for supplying and tuning the following parameters:

subset, na.action, ntree, mtry, replace, classwt, cutoff, strata, sampsize, nodesize, maxnodes

tunePareto.NaiveBayes encapsulates the NaiveBayes classifier in package klaR. The classifier allows for supplying and tuning the following parameters:

prior, usekernel, fL, subset, na.action, bw, adjust, kernel, weights, window, width, give.Rkern, n, from, to, cut, na.rm

Examples

Run this code

# \donttest{
# tune a k-NN classifier with different 'k' and 'l' 
# on the 'iris' data set
print(tunePareto(classifier = tunePareto.knn(),
                 data = iris[, -ncol(iris)], 
                 labels = iris[, ncol(iris)],
                 k = c(5,7,9),
                 l = c(1,2,3),
                 objectiveFunctions=list(cvError(10, 10),
                                         cvSpecificity(10, 10, caseClass="setosa"))))
                 
# tune an SVM with different costs on 
# the 'iris' data set
# using Halton sequences for sampling
print(tunePareto(classifier = tunePareto.svm(),
                 data = iris[, -ncol(iris)], 
                 labels = iris[, ncol(iris)],
                 cost = as.interval(0.001,10),
                 sampleType = "halton",
                 numCombinations=20,                 
                 objectiveFunctions=list(cvWeightedError(10, 10),
                                         cvSensitivity(10, 10, caseClass="setosa"))))

# tune a CART classifier with different 
# splitting criteria on the 'iris' data set
print(tunePareto(classifier = tunePareto.tree(),
                 data = iris[, -ncol(iris)], 
                 labels = iris[, ncol(iris)],
                 split = c("deviance","gini"),
                 objectiveFunctions=list(cvError(10, 10),
                                         cvErrorVariance(10, 10))))

# tune a Random Forest with different numbers of trees 
# on the 'iris' data set
print(tunePareto(classifier = tunePareto.randomForest(),
                 data = iris[, -ncol(iris)], 
                 labels = iris[, ncol(iris)],
                 ntree = seq(50,300,50),
                 objectiveFunctions=list(cvError(10, 10),
                                         cvSpecificity(10, 10, caseClass="setosa"))))

# tune a Naive Bayes classifier with different kernels
# on the 'iris' data set
print(tunePareto(classifier = tunePareto.NaiveBayes(),
                 data = iris[, -ncol(iris)], 
                 labels = iris[, ncol(iris)],
                 kernel = c("gaussian", "epanechnikov", "rectangular",
                            "triangular", "biweight",
                            "cosine", "optcosine"),
                 objectiveFunctions=list(cvError(10, 10),
                                         cvSpecificity(10, 10, caseClass="setosa"))))
                             
# }