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
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
|
\docType{methods}
\name{optimizing}
\alias{optimizing}
\alias{optimizing,stanmodel-method}
\title{Obtain a point estimate by maximizing the joint posterior}
\description{
Obtain a point estimate by maximizing the joint posterior
from the model defined by class \code{stanmodel}.
}
\usage{
%% optimizing(object, \dots)
\S4method{optimizing}{stanmodel}(object, data = list(),
seed = sample.int(.Machine$integer.max, 1), init = 'random',
check_data = TRUE, sample_file = NULL,
algorithm = c("LBFGS", "BFGS", "Newton"),
verbose = FALSE, hessian = FALSE, as_vector = TRUE,
draws = 0, constrained = TRUE, importance_resampling = FALSE, \dots)
}
\section{Methods}{
\describe{
\item{optimizing}{\code{signature(object = "stanmodel")}
Call Stan's optimization methods to obtain a point estimate
for the model defined by S4 class \code{stanmodel}
given the data, initial values, etc.
}
}
}
\arguments{
\item{object}{An object of class \code{\linkS4class{stanmodel}}.}
\item{data}{A named \code{list} or \code{environment}
providing the data for the model or a character vector
for all the names of objects used as data.
See the \strong{Passing data to Stan} section in \code{\link{stan}}.}
\item{seed}{The seed for random number generation. The default is generated
from 1 to the maximum integer supported by \R on the machine. Even if
multiple chains are used, only one seed is needed, with other chains having
seeds derived from that of the first chain to avoid dependent samples.
When a seed is specified by a number, \code{as.integer} will be applied to it.
If \code{as.integer} produces \code{NA}, the seed is generated randomly.
The seed can also be specified as a character string of digits, such as
\code{"12345"}, which is converted to integer.}
\item{init}{Initial values specification. See the detailed documentation for
the \code{init} argument in \code{\link{stan}} with one exception. If specifying inits
using a list then only a single named list of values should be provided.
For example, to initialize a parameter \code{alpha} to \code{value1} and
\code{beta} to \code{value2} you can specify \code{list(alpha = value1, beta = value2)}.}
\item{check_data}{Logical, defaulting to \code{TRUE}. If \code{TRUE}
the data will be preprocessed; otherwise not.
See the \strong{Passing data to Stan} section in \code{\link{stan}}.}
\item{sample_file}{A character string of file name for specifying where to
write samples for \emph{all} parameters and other saved quantities.
If not provided, files are not created. When the folder specified
is not writable, \code{tempdir()} is used.}
\item{algorithm}{One of \code{"Newton"}, \code{"BFGS"},
and \code{"LBFGS"} (the default) indicating which optimization algorithm
to use.}
\item{verbose}{\code{TRUE} or \code{FALSE} (the default): flag indicating
whether to print intermediate output from Stan on the console, which might
be helpful for model debugging.}
\item{hessian}{\code{TRUE} or \code{FALSE} (the default): flag indicating
whether to calculate the Hessian (via numeric differentiation of the
gradient function in the unconstrained parameter space).}
\item{as_vector}{\code{TRUE} (the default) or \code{FALSE}: flag indicating
whether a vector is used to store the point estimate found. A list can be
used instead by specifying it to be \code{FALSE}.}
\item{draws}{A non-negative integer (that defaults to zero) indicating how
many times to draw from a multivariate normal distribution whose parameters
are the mean vector and the inverse negative Hessian in the unconstrained
space. If \code{draws > 0} and \code{importance_resampling=TRUE}
then \code{log_p} and \code{log_g} will be computed and returned
(see description in the \strong{Value} section).}
\item{constrained}{A logical scalar indicating, if \code{draws > 0}, whether
the draws should be transformed to the constrained space defined in the
parameters block of the Stan program. Defaults to \code{TRUE}.}
\item{importance_resampling}{A logical scalar (defaulting to \code{FALSE})
indicating whether to do importance resampling to compute diagnostics on the
draws from the normal approximation to the posterior distribution.
If \code{TRUE} and \code{draws > 0} then \code{log_p}
and \code{log_g} will be computed and returned (see description in the
\strong{Value} section).}
\item{\dots}{Other optional parameters:
\itemize{
\item \code{iter} (\code{integer}), the maximum number of iterations,
defaulting to 2000.
\item \code{save_iterations} (logical), a flag indicating whether to save
the iterations, defaulting to \code{FALSE}.
\item \code{refresh} (\code{integer}), the number of interations between
screen updates, defaulting to 100.
\item \code{init_alpha} (\code{double}), for BFGS and LBFGS,
the line search step size for first iteration, defaulting to 0.001.
\item \code{tol_obj} (\code{double}), for BFGS and LBFGS,
the convergence tolerance on changes in objective function value,
defaulting to 1e-12.
\item \code{tol_rel_obj} (\code{double}), for BFGS and LBFGS,
the convergence tolerance on relative changes in objective function value,
defaulting to 1e4.
\item \code{tol_grad} (\code{double}), for BFGS and LBFGS,
the convergence tolerance on the norm of the gradient, defaulting to 1e-8.
\item \code{tol_rel_grad} (\code{double}), for BFGS and LBFGS,
the convergence tolerance on the relative norm of the gradient, defaulting
to 1e7.
\item \code{tol_param} (\code{double}), for BFGS and LBFGS,
the convergence tolerance on changes in parameter value, defaulting to 1e-8.
\item \code{history_size} (\code{integer}), for LBFGS,
the number of update vectors to use in Hessian approximations,
defaulting to 5.
}
Refer to the manuals for both CmdStan and Stan for more details.
}
}
\value{
A list with components:
\item{par}{The point estimate found. Its form (vector or list)
is determined by the \code{as_vector} argument.}
\item{value}{The value of the log-posterior (up to an additive constant,
the \code{"lp__"} in Stan) corresponding to \code{par}.}
\item{return_code}{The value of the return code from the optimizer;
anything that is not zero is problematic.}
\item{hessian}{The Hessian matrix if \code{hessian} is \code{TRUE}}
\item{theta_tilde}{If \code{draws > 0}, the matrix of parameter draws
in the constrained or unconstrained space, depending on the value of
the \code{constrained} argument.}
\item{log_p}{If \code{draws > 0} and \code{importance_resampling=TRUE},
a vector of length \code{draws} that contains the value of the
log-posterior evaluated at each row of \code{theta_tilde}.}
\item{log_g}{If \code{draws > 0}, a vector of length \code{draws} that
contains the value of the logarithm of the multivariate normal density
evaluated at each row of \code{theta_tilde}.}
If the optimization is not completed for reasons such as feeding wrong data,
it returns \code{NULL}.
}
\seealso{
\code{\linkS4class{stanmodel}}
}
\examples{\dontrun{
m <- stan_model(model_code = 'parameters {real y;} model {y ~ normal(0,1);}')
f <- optimizing(m, hessian = TRUE)
}}
|
