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
|
% Generated by roxygen2: do not edit by hand
% Please edit documentation in R/project.R
\name{project}
\alias{project}
\title{Projection to submodels}
\usage{
project(
object,
nterms = NULL,
solution_terms = NULL,
cv_search = TRUE,
ndraws = 400,
nclusters = NULL,
intercept = NULL,
seed = NULL,
regul = 1e-04,
...
)
}
\arguments{
\item{object}{Either a \code{refmodel}-type object created by
\link[=get_refmodel]{get_refmodel} or \link[=init_refmodel]{init_refmodel},
or an object which can be converted to a reference model using
\link[=get_refmodel]{get_refmodel}.}
\item{nterms}{Number of terms in the submodel (the variable combination is
taken from the \code{varsel} information). If a list, then the projection
is performed for each model size. Default is the model size suggested by
the variable selection (see function \code{suggest_size}). Ignored if
\code{solution_terms} is specified.}
\item{solution_terms}{Variable indices onto which the projection is done. If
specified, \code{nterms} is ignored.}
\item{cv_search}{If TRUE, then the projected coefficients after L1-selection
are computed without any penalization (or using only the regularization
determined by \code{regul}). If FALSE, then the coefficients are the
solution from the L1-penalized projection. This option is relevant only if
L1-search was used. Default is TRUE for genuine reference models and FALSE
if \code{object} is datafit (see \link[=init_refmodel]{init_refmodel}).}
\item{ndraws}{Number of posterior draws to be projected. Ignored if
\code{nclusters} is specified. Default is 400.}
\item{nclusters}{Number of clusters in the clustered projection.}
\item{intercept}{Whether to use intercept. Default is \code{TRUE}.}
\item{seed}{A seed used in the clustering (if \code{nclusters!=ndraws}). Can
be used to ensure same results every time. @param regul Amount of
regularization in the projection. Usually there is no need for
regularization, but sometimes for some models the projection can be
ill-behaved and we need to add some regularization to avoid numerical
problems.}
\item{regul}{Ridgre regularization constant to fit the projections.}
\item{...}{Currently ignored.}
}
\value{
A list of submodels (or a single submodel if projection was
performed onto a single variable combination), each of which contains the
following elements:
\describe{
\item{\code{kl}}{The KL divergence from the reference model to the
submodel.} \item{\code{weights}}{Weights for each draw of the projected
model.}
\item{\code{dis}}{Draws from the projected dispersion parameter.}
\item{\code{alpha}}{Draws from the projected intercept.}
\item{\code{beta}}{Draws from the projected weight vector.}
\item{\code{solution_terms}}{The order in which the variables were added to
the submodel.}
\item{\code{intercept}}{Whether or not the model contains an
intercept.}
\item{\code{family}}{A modified \code{\link{family}}-object.}
}
}
\description{
Perform projection onto submodels of selected sizes or a specified feature
combination.
}
\examples{
\donttest{
if (requireNamespace("rstanarm", quietly = TRUE)) {
### Usage with stanreg objects
n <- 30
d <- 5
x <- matrix(rnorm(n * d), nrow = n)
y <- x[, 1] + 0.5 * rnorm(n)
data <- data.frame(x, y)
fit <- rstanarm::stan_glm(y ~ X1 + X2 + X3 + X4 + X5, gaussian(),
data = data, chains = 2, iter = 500)
vs <- varsel(fit)
# project onto the best model with 4 variables
proj4 <- project(vs, nterms = 4)
# project onto an arbitrary variable combination (variable indices 1, 3 and 5)
proj <- project(fit, solution_terms = c(1, 3, 5))
}
}
}
|
