Tests on artificial data with 100% noise

library(adabag)
library(naivebayes)

Loading required package: rpart
Loading required package: caret
Loading required package: lattice
Loading required package: ggplot2
Loading required package: foreach
Loading required package: doParallel
Loading required package: iterators
Loading required package: parallel

# files available in: /net/aistaff/kleiweg/spraak/fa
train = read.table("data100.train", header=TRUE, sep="\t", quote="", row.names=1)
test  = read.table("data100.test",  header=TRUE, sep="\t", quote="", row.names=1)

train[1:10,]

	C.Class	C.W1	C.W2	C.W3	C.W4	C.W5	C.W6	C.W7	C.W8	C.W9	⋯	C.W11	C.W12	C.W13	C.W14	C.W15	C.W16	C.W17	C.W18	C.W19	C.W20
921	B	1.B1	2.B1	3.C1	4.C1	5.B1	6.A2	7.A1	8.B4	9.A1	⋯	11.A1	12.B3	13.A1	14.C1	15.C5	16.A1	17.A2	18.A1	19.C3	20.B3
470	B	1.C4	2.B3	3.B3	4.B1	5.C1	6.A1	7.B1	8.C1	9.A1	⋯	11.B3	12.C1	13.B5	14.B2	15.C4	16.B1	17.B1	18.C1	19.A1	20.A1
204	C	1.B4	2.B3	3.B3	4.B1	5.B3	6.C1	7.C2	8.A1	9.A1	⋯	11.B4	12.C1	13.A1	14.B3	15.B2	16.A1	17.B1	18.B1	19.A1	20.A1
333	C	1.B2	2.B4	3.B3	4.A1	5.A2	6.B1	7.A3	8.A2	9.A1	⋯	11.B3	12.C1	13.C4	14.C1	15.C3	16.B1	17.C8	18.A1	19.B3	20.C2
131	B	1.A1	2.A1	3.A4	4.C1	5.C1	6.B1	7.A3	8.A2	9.B3	⋯	11.B2	12.A8	13.B2	14.A2	15.A1	16.A1	17.A2	18.B1	19.C3	20.A1
958	C	1.B4	2.A1	3.C1	4.C1	5.C1	6.B1	7.C3	8.B4	9.A2	⋯	11.C3	12.B1	13.C1	14.B3	15.C5	16.B1	17.B1	18.A1	19.A1	20.C1
866	B	1.C2	2.A1	3.B1	4.B1	5.B1	6.B1	7.B1	8.A1	9.A1	⋯	11.B4	12.A8	13.B2	14.C1	15.B1	16.A1	17.A4	18.B1	19.A1	20.B1
952	B	1.C2	2.C1	3.B1	4.B1	5.B3	6.C1	7.A3	8.C1	9.B3	⋯	11.C4	12.C1	13.B4	14.C1	15.B3	16.A1	17.C6	18.B1	19.B1	20.B2
216	C	1.C2	2.C1	3.C5	4.A1	5.B2	6.A1	7.A3	8.C1	9.A1	⋯	11.A1	12.A4	13.A1	14.A2	15.B2	16.C2	17.B1	18.A1	19.A1	20.B1
325	C	1.C1	2.C1	3.A5	4.A1	5.B2	6.B1	7.C3	8.A1	9.B2	⋯	11.C4	12.C1	13.C4	14.A1	15.A1	16.A1	17.C1	18.B1	19.C2	20.C2

Bagging (AdaBag)

bag <- bagging(C.Class ~ ., data=train)
train.bagging <- predict(bag, newdata=train)
 test.bagging <- predict(bag, newdata=test)
100 * (1 - train.bagging$error)
100 * (1 -  test.bagging$error)

86

33

Boosting (AdaBoost)

boost <- boosting(C.Class ~ ., data=train)
train.boosting <- predict(boost, newdata=train)
 test.boosting <- predict(boost, newdata=test)
100 * (1 - train.boosting$error)
100 * (1 -  test.boosting$error)

100

39

Naive Bayes

score <- function(obs, exp) {
  return(100 * sum(obs == exp[,"C.Class"]) / length(obs))
}

nb <- naive_bayes(C.Class ~ ., data=train)
train.nb <- predict(nb, train)
 test.nb <- predict(nb, test)
score(train.nb, train)
score( test.nb, test)

56.7777777777778

32

simpel.go

out <- system2(c("./simpel", "data100.train", "data100.test"), stdout=TRUE, stderr=TRUE)
cat(out, sep="\n")

Training score:	 46.9%
Testing score:	 31.0%