1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
|
\name{sample.split}
\alias{sample.split}
\title{Split Data into Test and Train Set}
\description{Split data from vector Y into two sets in predefined ratio while
preserving relative ratios of different labels in Y. Used to split the data
used during classification into train and test subsets.
}
\usage{
sample.split( Y, SplitRatio = 2/3, group = NULL )
}
\arguments{
\item{Y}{Vector of data labels. If there are only a few labels (as is
expected) than relative ratio of data in both subsets will be the same.}
\item{SplitRatio}{ Splitting ratio:
\itemize{
\item if \code{(0<=SplitRatio<1)} then \code{SplitRatio} fraction of
points from Y will be set toTRUE
\item if \code{(SplitRatio==1)} then one random point from Y will be set
to TRUE
\item if \code{(SplitRatio>1)} then \code{SplitRatio} number of points
from Y will be set to TRUE
}
}
\item{group}{Optional vector/list used when multiple copies of each sample
are present. In such a case \code{group} contains unique sample labels,
marking all copies of the same sample with the same
label, and the function tries to place all copies in either train or test
subset. If provided than has to have the same length as \code{Y}.}
}
\details{ Function \code{msc.sample.split} is the old name of the
\code{sample.split} function. To be retired soon.
}
\value{Returns logical vector of the same length as Y with random
\code{SplitRatio*length(Y)} elements set to TRUE.
}
\author{Jarek Tuszynski (SAIC) \email{jaroslaw.w.tuszynski@saic.com}}
\seealso{
\itemize{
\item Similar to \code{\link{sample}} function.
\item Variable \code{group} is used in the same way as \code{f} argument in
\code{\link{split}} and \code{INDEX} argument in \code{\link{tapply}}
}
}
\examples{
library(MASS)
data(cats) # load cats data
Y = cats[,1] # extract labels from the data
msk = sample.split(Y, SplitRatio=3/4)
table(Y,msk)
t=sum( msk) # number of elements in one class
f=sum(!msk) # number of elements in the other class
stopifnot( round((t+f)*3/4) == t ) # test ratios
# example of using group variable
g = rep(seq(length(Y)/4), each=4); g[48]=12;
msk = sample.split(Y, SplitRatio=1/2, group=g)
table(Y,msk) # try to get correct split ratios ...
split(msk,g) # ... while keeping samples with the same group label together
# test results
print(paste( "All Labels numbers: total=",t+f,", train=",t,", test=",f,
", ratio=", t/(t+f) ) )
U = unique(Y) # extract all unique labels
for( i in 1:length(U)) { # check for all labels
lab = (Y==U[i]) # mask elements that have label U[i]
t=sum( msk[lab]) # number of elements with label U[i] in one class
f=sum(!msk[lab]) # number of elements with label U[i] in the other class
print(paste( "Label",U[i],"numbers: total=",t+f,", train=",t,", test=",f,
", ratio=", t/(t+f) ) )
}
# use results
train = cats[ msk,2:3] # use output of sample.split to ...
test = cats[!msk,2:3] # create train and test subsets
z = lda(train, Y[msk]) # perform classification
table(predict(z, test)$class, Y[!msk]) # predicted & true labels
# see also LogitBoost example
}
\keyword{classif}
|
