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
|
% Generated by roxygen2: do not edit by hand
% Please edit documentation in R/importance_sampling.R
\name{importance_sampling}
\alias{importance_sampling}
\alias{importance_sampling.array}
\alias{importance_sampling.matrix}
\alias{importance_sampling.default}
\title{A parent class for different importance sampling methods.}
\usage{
importance_sampling(log_ratios, method, ...)
\method{importance_sampling}{array}(
log_ratios,
method,
...,
r_eff = 1,
cores = getOption("mc.cores", 1)
)
\method{importance_sampling}{matrix}(
log_ratios,
method,
...,
r_eff = 1,
cores = getOption("mc.cores", 1)
)
\method{importance_sampling}{default}(log_ratios, method, ..., r_eff = 1)
}
\arguments{
\item{log_ratios}{An array, matrix, or vector of importance ratios on the log
scale (for PSIS-LOO these are \emph{negative} log-likelihood values). See the
\strong{Methods (by class)} section below for a detailed description of how
to specify the inputs for each method.}
\item{method}{The importance sampling method to use. The following methods
are implemented:
\itemize{
\item \code{\link[=psis]{"psis"}}: Pareto-Smoothed Importance Sampling (PSIS). Default method.
\item \code{\link[=tis]{"tis"}}: Truncated Importance Sampling (TIS) with truncation at
\code{sqrt(S)}, where \code{S} is the number of posterior draws.
\item \code{\link[=sis]{"sis"}}: Standard Importance Sampling (SIS).
}}
\item{...}{Arguments passed on to the various methods.}
\item{r_eff}{Vector of relative effective sample size estimates containing
one element per observation. The values provided should be the relative
effective sample sizes of \code{1/exp(log_ratios)} (i.e., \code{1/ratios}).
This is related to the relative efficiency of estimating the normalizing
term in self-normalizing importance sampling. If \code{r_eff} is not
provided then the reported PSIS effective sample sizes and Monte Carlo
error estimates can be over-optimistic. If the posterior draws are (near)
independent then \code{r_eff=1} can be used. \code{r_eff} has to be a scalar (same
value is used for all observations) or a vector with length equal to the
number of observations. The default value is 1. See the \code{\link[=relative_eff]{relative_eff()}}
helper function for computing \code{r_eff}.}
\item{cores}{The number of cores to use for parallelization. This defaults to
the option \code{mc.cores} which can be set for an entire R session by
\code{options(mc.cores = NUMBER)}. The old option \code{loo.cores} is now
deprecated but will be given precedence over \code{mc.cores} until
\code{loo.cores} is removed in a future release. \strong{As of version
2.0.0 the default is now 1 core if \code{mc.cores} is not set}, but we
recommend using as many (or close to as many) cores as possible.
\itemize{
\item Note for Windows 10 users: it is \strong{strongly}
\href{https://github.com/stan-dev/loo/issues/94}{recommended} to avoid using
the \code{.Rprofile} file to set \code{mc.cores} (using the \code{cores} argument or
setting \code{mc.cores} interactively or in a script is fine).
}}
}
\description{
A parent class for different importance sampling methods.
}
|
