# R code: classification examples

Cross-validation for logistic regression:

library(languageR)

library(rms)

library(DAAG)

m.lrm = lrm(RealizationOfRecipient ~ Modality + SemanticClass +

LengthOfRecipient + AnimacyOfRec + DefinOfRec + PronomOfRec +

LengthOfTheme + AnimacyOfTheme + DefinOfTheme + PronomOfTheme +

AccessOfRec + AccessOfTheme, data = dative)

# cross-validation

# first estimate the regression model using glm rather than lrm

m.glm = glm(RealizationOfRecipient ~ Modality + SemanticClass +

LengthOfRecipient + AnimacyOfRec + DefinOfRec + PronomOfRec +

LengthOfTheme + AnimacyOfTheme + DefinOfTheme + PronomOfTheme +

AccessOfRec + AccessOfTheme, data = dative, family="binomial")

summary(m.glm)

CVbinary(m.glm)

Support vector machines:

library(e1071)

# SVM can only deal with numeric predictors

d = dative

for (i in 2:12) d[,i] = as.numeric(d[,i])

for (i in 14:15) d[,i] = as.numeric(d[,i])

m.svm = svm(d[c(2:12,14:15)], d$RealizationOfRecipient, cross=10)

summary(m.svm)

# or:

d.predictors = d[c(2:12,14:15)]

tune(svm, d.predictors, dative$RealizationOfRecipient)

A Naive Bayes classifier:

library(e1071)

# for Naive Bayes, we want to use categorial predictors where we can,

# as for them the output is more informative

m.nb = naiveBayes(dative[,c(2:12,14:15)],

dative$RealizationOfRecipient)

# unfortunately, e1071's "tune" doesn't work for Naive Bayes.

# so we do this by hand

d2 = dative

d2$fold = cut(1:nrow(d2), breaks=10, labels=F)

# this gives the following folds:

unique(d2$fold)

accuracies = c()

for (i in 1:10) {

m.nbi = naiveBayes(d2[d2$fold != i,c(2:12,14:15)], d2[d2$fold != i,]$RealizationOfRecipient)

predictions = predict(m.nbi, d2[d2$fold == i, c(2:12, 14:15)])

numcorrect = sum(predictions == d2[d2$fold == i,]$RealizationOfRecipient)

accuracies = append(numcorrect / nrow(d2[d2$fold == i,]), accuracies)

}

accuracies

mean(accuracies)

Using the same cross-validation technique for other classifiers:

# we can do the same for the SVM

d$fold = cut(1:nrow(d), breaks=10, labels=F)

accuracies = c()

for (i in 1:10) {

m.svmi = svm(d[d$fold != i,c(2:12,14:15)], d[d$fold != i,]$RealizationOfRecipient)

predictions = predict(m.svmi, d[d$fold == i, c(2:12, 14:15)])

numcorrect = sum(predictions == d[d$fold == i,]$RealizationOfRecipient)

accuracies = append(numcorrect / nrow(d[d$fold == i,]), accuracies)

}

accuracies

mean(accuracies)

# and for logistic regression

invlogit = function(x) { 1/(1+exp(-x)) }

d2.predictors = d2[,c(2, 4:12,14:15)]

accuracies = c()

for (i in 1:10) {

m.li = lrm(RealizationOfRecipient ~ Modality + SemanticClass +

LengthOfRecipient + AnimacyOfRec + DefinOfRec + PronomOfRec +

LengthOfTheme + AnimacyOfTheme + DefinOfTheme + PronomOfTheme +

AccessOfRec + AccessOfTheme,

data = d2[d2$fold != i,])

predictions = round(invlogit(predict(m.li, d2.predictors[d2$fold ==

i,])))

actual = as.numeric(d2[d2$fold == i,]$RealizationOfRecipient) - 1

numcorrect = sum(predictions == actual)

accuracies = append(numcorrect / nrow(d2[d2$fold == i,]), accuracies)

}

accuracies

mean(accuracies)