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
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
|
\name{stab_control}
\alias{stab_control}
\title{Control for Supervised Stability Assessments}
\description{
Various parameters that control aspects of the stability assessment performed
via \code{\link{stability}}.
}
\usage{
stab_control(B = 500, measure = list(tvdist, ccc), sampler = "bootstrap",
evaluate = "OOB", holdout = 0.25, seed = NULL, na.action = na.exclude,
savepred = TRUE, silent = TRUE, ...)
}
\arguments{
\item{B}{an integer value specifying the number of repetitions. The default
is \code{B = 500}.}
\item{measure}{a list of similarity measure (generating) functions. Those
should either be functions of \code{p1} and \code{p2} (the predictions of
two results in a repetition, see Details below) or similarity measure
generator functions, see \code{\link{similarity_measures_classification}}
and \code{\link{similarity_measures_regression}}. The defaults are
\code{\link{tvdist}} for the classification and \code{\link{ccc}} for the
regression case.}
\item{sampler}{a resampling (generating) function. Either this should be a
function of \code{n} that returns a matrix or a sampler generator like
\code{\link{bootstrap}}. The default is \code{"bootstrap"}.}
\item{evaluate}{a character specifying the evaluation strategy to be applied
(see Details below). The default is \code{"OOB"}.}
\item{holdout}{a numeric value between zero and one that specifies the
proportion of observations hold out for evaluation over all repetitions,
only if \code{evaluate = "OOB"}. The default is \code{0.25}.}
\item{seed}{a single value, interpreted as an integer, see
\code{\link{set.seed}} and Details section below. The default is \code{NULL},
which indicates that no seed has been set.}
\item{na.action}{a function which indicates what should happen when the
predictions of the results contain \code{NAs}. The default function is
\code{\link{na.exclude}}.}
\item{savepred}{logical. Should the predictions from each iteration be
saved? If \code{savepred = FALSE}, the resulting object will be smaller.
However, predictions are required to subsequently compute measures of
accuracy via \code{\link{accuracy}}.}
\item{silent}{logical. If \code{TRUE}, error messages, generated by the
learner while training or predicting, are suppressed and \code{NA} is
returned for the corresponding iteration of the stability assessment
procedure.}
\item{\dots}{arguments passed to \code{sampler} function.}
}
\details{
With the argument \code{measure} one or more measures can be defined that are
used to assess the stability of a result from supervised statistical learning
by \code{\link{stability}}. Predefined similarity measures for the regression
and the classification case are listed in \code{\link{similarity_measures_classification}}
and \code{\link{similarity_measures_regression}}.
Users can define their own similarity functions \code{f(p1, p2)} that must
return a single numeric value for the similarity between two results trained on
resampled data sets. Such a function must take the arguments \code{p1} and \code{p2}.
In the classification case, \code{p1} and \code{p2} are probability matrices of
size m * K, where \code{m} is the number of predicted observations (size
of the evaluation sample) and K is the number of classes. In the
regression case, \code{p1} and \code{p2} are numeric vectors of length
m.
A different way to implement new similarity functions for the current \R
session is to define a similarity measure generator function, which is a
function without arguments that generates a list of five elements including the
name of the similarity measure, the function to compute the similarity
between the predictions as described above, a vector of character values
specifying the response types for which the similarity measure can be used,
a list containing two numeric elements \code{lower} and \code{upper} that
specify the range of values of the similarity measure and the function to
invert (or reverse) the similarity values such that higher values indicate
higher stability. The latter can be set to \code{NULL}, if higher similarity
values already indicate higher stability. Those elements should be named
\code{name}, \code{measure}, \code{classes}, \code{range} and \code{reverse}.
The argument \code{evaluate} can be used to specify the evaluation strategy.
If set to \code{"ALL"}, all observations in the original data set are used for
evaluation. If set to \code{"OOB"}, only the pairwise out-of-bag observations
are used for evaluation within each repetition. If set to \code{"OOS"}, a
fraction (defined by \code{holdout}) of the observations in the original data
set are randomly sampled and used for evaluation, but not for training, over all
repetitions.
The argument \code{seed} can be used to make similarity assessments comparable
when comparing the stability of different results that were trained on the same
data set. By default, \code{seed} is set to \code{NULL} and the learning samples
are sampled independently for each fitted model object passed to
\code{\link{stability}}. If \code{seed} is set to a specific number, the seed
will be set for each fitted model object before the learning samples are
generated using \code{"L'Ecuyer-CMRG"} (see \code{\link{set.seed}}) which
guarantees identical learning samples for each stability assessment and, thus,
comparability of the stability assessments between the results.}
\seealso{\code{\link{stability}}}
\examples{
\donttest{
library("partykit")
res <- ctree(Species ~ ., data = iris)
## less repetitions
stability(res, control = stab_control(B = 100))
}
\dontrun{
## change similarity measure
stability(res, control = stab_control(measure = list(bdist)))
## change evaluation strategy
stability(res, control = stab_control(evaluate = "ALL"))
stability(res, control = stab_control(evaluate = "OOS"))
## change resampling strategy to subsampling
stability(res, control = stab_control(sampler = subsampling))
stability(res, control = stab_control(sampler = subsampling, evaluate = "ALL"))
stability(res, control = stab_control(sampler = subsampling, evaluate = "OOS"))
## change resampling strategy to splithalf
stability(res, control = stab_control(sampler = splithalf, evaluate = "ALL"))
stability(res, control = stab_control(sampler = splithalf, evaluate = "OOS"))
}
}
\keyword{resampling}
\keyword{similarity}
|
