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
|
\name{predict.earth}
\alias{predict.earth}
\title{Predict with an earth model}
\description{
Predict with an \code{\link{earth}} model.
}
\usage{
\method{predict}{earth}(object = stop("no 'object' argument"), newdata = NULL,
type = c("link", "response", "earth", "class", "terms"),
interval = "none", level = .95,
thresh = .5, trace = FALSE, \dots)
}
\arguments{
\item{object}{
An \code{\link{earth}} object.
This is the only required argument.
}
\item{newdata}{
Make predictions using \code{newdata}, which
can be a data frame, a matrix, or a vector with length equal to a
multiple of the number of columns
of the original input matrix \code{x}.\cr
Default is NULL, meaning return values predicted from the training set.\cr
NAs are allowed in \code{newdata} (and the predicted value will be NA
unless the NAs are in variables that are unused in the earth model).
}
\item{type}{
Type of prediction.
One of \code{"link"} (default), \code{"response"}, \code{"earth"}, \code{"class"}, or \code{"terms"}.
See the \bold{Note} below.
}
\item{interval}{
Return prediction or confidence levels.
Default is \code{"none"}.
Use \code{interval="pint"} to get prediction intervals on new data.
\cr
Requires that the earth model was built with \code{varmod.method}.
\cr
This argument gets passed on as the \code{type} argument to \code{\link{predict.varmod}}.
See its help page for details.
}
\item{level}{
Confidence level for the \code{interval} argument.
Default is \code{0.95}, meaning construct 95\% confidence bands
(estimate the 2.5\% and 97.5\% levels).
}
\item{thresh}{
Threshold, a value between 0 and 1 when predicting a probability.
Only applies when \code{type="class"}.
Default is 0.5.
See the \bold{Note} below.
}
\item{trace}{
Default \code{FALSE}. Set to \code{TRUE} to see which data, subset, etc. \code{predict.earth} is using.
}
\item{\dots}{
Unused, but provided for generic/method consistency.
}
}
\value{
The predicted values (a matrix for multiple response models).
If \code{type="terms"}, a matrix with each column showing the contribution of a predictor.
If \code{interval="pint"} or \code{"cint"}, a matrix with three columns:\cr
\code{fit}: the predicted values\cr
\code{lwr}: the lower confidence or prediction limit\cr
\code{upr}: the upper confidence or prediction limit
If \code{interval="se"}, the standard errors.
}
\note{
\bold{Predicting with standard earth models}
Use the default \code{type="link"}, or possibly \code{type="class"}.
Actually, the \code{"link"}, \code{"response"}, and \code{"earth"}
choices all return the same value unless the \code{glm} argument
was used in the original call to \code{\link{earth}}.
\bold{Predicting with earth-GLM models}
This section applies to earth models with a GLM component, i.e.,
when the \code{glm} argument was used
in the original call to \code{\link{earth}}.
The \code{"link"} and \code{"response"} options:
see \code{\link{predict.glm}} for a description of these.
In brief: for logistic models
use \code{type="response"} to get probabilities,
and \code{type="link"} to get log-odds.
Use option \code{"earth"} to get the linear fit (this gives the prediction you would get
if your original call to earth had no \code{glm} argument).
\bold{Predicting with "class"}
Use option \code{"class"} to get the predicted class.
With option \code{"class"}, this function first makes predictions with
\code{type="response"} and then assigns the predicted values to classes as follows:
(i) When the response is a \emph{logical}, predict \code{TRUE} if
the predicted probability is greater than \code{thresh} (default \code{0.5}).
(ii) When the response is a \emph{numeric}, predict \code{TRUE} if
the predicted value is greater than \code{thresh}.
Actually, this is identical to the above case,
although \code{thresh} here may legitimately be a value
outside the 0...1 range.
(iii) When the response is a \emph{two level factor},
predict the second level if its probability is more than \code{thresh}.
In other words, with the default \code{thresh=0.5} predict the most probable level.
(iv) When the response is a \emph{three or more level factor},
predict the most probable level (and \code{thresh} is ignored).
\bold{Predicting with "terms"}
The \code{"terms"} option returns a \code{"link"} response suitable for \code{\link{termplot}}.
Only the additive terms and the first response (for multi-response models) are returned.
Also, \code{"terms"} always returns the earth terms, and ignores the GLM component
of the model, if any.
}
\seealso{
\code{\link{earth}},
\code{\link{predict}}
}
\examples{
data(trees)
earth.mod <- earth(Volume ~ ., data = trees)
predict(earth.mod) # same as earth.mod$fitted.values
predict(earth.mod, c(10,80)) # yields 16.8
}
\keyword{models}
|
