-
Notifications
You must be signed in to change notification settings - Fork 19
/
Copy pathstep_pca_sparse_bayes.Rd
175 lines (151 loc) · 6.59 KB
/
step_pca_sparse_bayes.Rd
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
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
% Generated by roxygen2: do not edit by hand
% Please edit documentation in R/pca_sparse_bayes.R
\name{step_pca_sparse_bayes}
\alias{step_pca_sparse_bayes}
\alias{tidy.step_pca_sparse_bayes}
\title{Sparse Bayesian PCA Signal Extraction}
\usage{
step_pca_sparse_bayes(
recipe,
...,
role = "predictor",
trained = FALSE,
num_comp = 5,
prior_slab_dispersion = 1,
prior_mixture_threshold = 0.1,
options = list(),
res = NULL,
prefix = "PC",
keep_original_cols = FALSE,
skip = FALSE,
id = rand_id("pca_sparse_bayes")
)
}
\arguments{
\item{recipe}{A recipe object. The step will be added to the sequence of
operations for this recipe.}
\item{...}{One or more selector functions to choose which variables will be
used to compute the components. See \link[recipes:selections]{recipes::selections} for more details. For
the \code{tidy} method, these are not currently used.}
\item{role}{For model terms created by this step, what analysis role should
they be assigned? By default, the function assumes that the new principal
component columns created by the original variables will be used as
predictors in a model.}
\item{trained}{A logical to indicate if the quantities for preprocessing have
been estimated.}
\item{num_comp}{The number of components to retain as new predictors.
If \code{num_comp} is greater than the number of columns or the number of
possible components, a smaller value will be used. If \code{num_comp = 0}
is set then no transformation is done and selected variables will
stay unchanged, regardless of the value of \code{keep_original_cols}.}
\item{prior_slab_dispersion}{This value is proportional to the dispersion (or
scale) parameter for the slab portion of the prior. Smaller values result
in an increase in zero coefficients.}
\item{prior_mixture_threshold}{The parameter that defines the trade-off
between the spike and slab components of the prior. Increasing this
parameter increases the number of zero coefficients.}
\item{options}{A list of options to the default method for
\code{\link[VBsparsePCA:VBsparsePCA]{VBsparsePCA::VBsparsePCA()}}.}
\item{res}{The rotation matrix once this preprocessing step has been trained
by \link[recipes:prep]{recipes::prep}.}
\item{prefix}{A character string that will be the prefix to the resulting new
variables. See notes below.}
\item{keep_original_cols}{A logical to keep the original variables in the
output. Defaults to \code{FALSE}.}
\item{skip}{A logical. Should the step be skipped when the recipe is baked by
\code{\link[recipes:bake]{recipes::bake()}}? While all operations are baked when \code{\link[recipes:prep]{recipes::prep()}} is
run, some operations may not be able to be conducted on new data (e.g.
processing the outcome variable(s)). Care should be taken when using \code{skip = TRUE} as it may affect the computations for subsequent operations}
\item{id}{A character string that is unique to this step to identify it.}
}
\value{
An updated version of \code{recipe} with the new step added to the
sequence of existing steps (if any). For the \code{tidy} method, a tibble with
columns \code{terms} (the selectors or variables selected), \code{value} (the
loading), and \code{component}.
}
\description{
\code{step_pca_sparse_bayes()} creates a \emph{specification} of a recipe step that
will convert numeric data into one or more principal components that can have
some zero coefficients.
}
\details{
The \code{VBsparsePCA} package is required for this step. If it is not installed,
the user will be prompted to do so when the step is defined.
A spike-and-slab prior is a mixture of two priors. One (the "spike") has all
of its mass at zero and represents a variable that has no contribution to the
PCA coefficients. The other prior is a broader distribution that reflects the
coefficient distribution of variables that do affect the PCA analysis. This
is the "slab". The narrower the slab, the more likely that a coefficient will
be zero (or are regularized to be closer to zero). The mixture of these two
priors is governed by a mixing parameter, which itself has a prior
distribution and a hyper-parameter prior.
PCA coefficients and their resulting scores are unique only up to the sign.
This step will attempt to make the sign of the components more consistent
from run-to-run. However, the sparsity constraint may interfere with this
goal.
The argument \code{num_comp} controls the number of components that will be retained
(the original variables that are used to derive the components are removed from
the data). The new components will have names that begin with \code{prefix} and a
sequence of numbers. The variable names are padded with zeros. For example, if
\code{num_comp < 10}, their names will be \code{PC1} - \code{PC9}. If \code{num_comp = 101},
the names would be \code{PC1} - \code{PC101}.
}
\section{Tidying}{
When you \code{\link[recipes:tidy.recipe]{tidy()}} this step, a tibble is returned with
columns \code{terms}, \code{value}, \code{component}, and \code{id}:
\describe{
\item{terms}{character, the selectors or variables selected}
\item{value}{numeric, variable loading}
\item{component}{character, principle component}
\item{id}{character, id of this step}
}
}
\section{Tuning Parameters}{
This step has 3 tuning parameters:
\itemize{
\item \code{num_comp}: # Components (type: integer, default: 5)
\item \code{prior_slab_dispersion}: Dispersion of Slab Prior (type: double, default: 1)
\item \code{prior_mixture_threshold}: Threshold for Mixture Prior (type: double, default: 0.1)
}
}
\section{Case weights}{
The underlying operation does not allow for case weights.
}
\examples{
\dontshow{if (rlang::is_installed(c("modeldata", "ggplot2"))) (if (getRversion() >= "3.4") withAutoprint else force)(\{ # examplesIf}
library(recipes)
library(ggplot2)
data(ad_data, package = "modeldata")
ad_rec <-
recipe(Class ~ ., data = ad_data) \%>\%
step_zv(all_predictors()) \%>\%
step_YeoJohnson(all_numeric_predictors()) \%>\%
step_normalize(all_numeric_predictors()) \%>\%
step_pca_sparse_bayes(
all_numeric_predictors(),
prior_mixture_threshold = 0.95,
prior_slab_dispersion = 0.05,
num_comp = 3,
id = "sparse bayesian pca"
) \%>\%
prep()
tidy(ad_rec, id = "sparse bayesian pca") \%>\%
mutate(value = ifelse(value == 0, NA, value)) \%>\%
ggplot(aes(x = component, y = terms, fill = value)) +
geom_tile() +
scale_fill_gradient2() +
theme(axis.text.y = element_blank())
\dontshow{\}) # examplesIf}
}
\references{
Ning, B. (2021). Spike and slab Bayesian sparse principal
component analysis. arXiv:2102.00305.
}
\seealso{
\code{\link[=step_pca_sparse]{step_pca_sparse()}}
}
\concept{pca}
\concept{preprocessing}
\concept{projection_methods}
\keyword{datagen}