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
|
% Generated by roxygen2: do not edit by hand
% Please edit documentation in R/tool_transformations_collapse.R
\name{plm.fast}
\alias{plm.fast}
\title{Option to Switch On/Off Fast Data Transformations}
\description{
A significant speed up can be gained by using fast (panel) data transformation
functions from package \code{collapse}.
An additional significant speed up for the two-way fixed effects case can be
achieved if package \code{fixest} or \code{lfe} is installed (package \code{collapse}
needs to be installed for the fast mode in any case).
}
\details{
By default, this speed up is enabled.
Option \code{plm.fast} can be used to enable/disable the speed up. The option is
evaluated prior to execution of supported transformations (see below), so
\code{option("plm.fast" = TRUE)} enables the speed up while
\code{option("plm.fast" = FALSE)} disables the speed up.
To have it always switched off, put \code{options("plm.fast" = FALSE)} in your
.Rprofile file.
See \strong{Examples} for how to use the option and for a benchmarking example.
For long, package \code{plm} used base R implementations and R-based code. The
package \code{collapse} provides fast data transformation functions written
in C/C++, among them some especially suitable for panel data.
Having package \code{collapse} installed is a requirement for the speed up, so
this package is a hard dependency for package \code{plm}.
Availability of packages \code{fixest} and \code{lfe} is checked for once when
package plm is attached and the additional speed up for the two-way fixed
effect case is enabled automatically (\code{fixest} wins over \code{lfe}),
given one of the packages is detected and \code{options("plm.fast" = TRUE)}
(default) is set. If so, the packages' fast algorithms to partial out fixed
effects are used (\code{fixest::demean} (via \code{collapse::fhdwithin}),
\code{lfe::demeanlist}). Both packages are 'Suggests' dependencies.
Users might experience neglectable numerical differences between enabled and
disabled fast mode and base R implementation, depending on the platform and
the additional packages installed.
Currently, these basic functions benefit from the speed-up, used as building
blocks in most model estimation functions, e.g., in \code{plm} (more functions are
under investigation):
\itemize{
\item between,
\item Between,
\item Sum,
\item Within,
\item pseriesfy.
}
}
\examples{
\dontrun{
### A benchmark of plm without and with speed-up
library("plm")
library("collapse")
library("microbenchmark")
rm(list = ls())
data("wlddev", package = "collapse")
form <- LIFEEX ~ PCGDP + GINI
# produce big data set (taken from collapse's vignette)
wlddevsmall <- get_vars(wlddev, c("iso3c","year","OECD","PCGDP","LIFEEX","GINI","ODA"))
wlddevsmall$iso3c <- as.character(wlddevsmall$iso3c)
data <- replicate(100, wlddevsmall, simplify = FALSE)
rm(wlddevsmall)
uniquify <- function(x, i) {
x$iso3c <- paste0(x$iso3c, i)
x
}
data <- unlist2d(Map(uniquify, data, as.list(1:100)), idcols = FALSE)
data <- pdata.frame(data, index = c("iso3c", "year"))
pdim(data) # Balanced Panel: n = 21600, T = 59, N = 1274400 // but many NAs
# data <- na.omit(data)
# pdim(data) # Unbalanced Panel: n = 13300, T = 1-31, N = 93900
times <- 1 # no. of repetitions for benchmark - this takes quite long!
onewayFE <- microbenchmark(
{options("plm.fast" = FALSE); plm(form, data = data, model = "within")},
{options("plm.fast" = TRUE); plm(form, data = data, model = "within")},
times = times)
summary(onewayFE, unit = "relative")
## two-ways FE benchmark requires pkg fixest and lfe
## (End-users shall only set option plm.fast. Option plm.fast.pkg.FE.tw shall
## _not_ be set by the end-user, it is determined automatically when pkg plm
## is attached; however, it needs to be set explicitly in this example for the
## benchmark.)
if(requireNamespace("fixest", quietly = TRUE) &&
requireNamespace("lfe", quietly = TRUE)) {
twowayFE <- microbenchmark(
{options("plm.fast" = FALSE);
plm(form, data = data, model = "within", effect = "twoways")},
{options("plm.fast" = TRUE, "plm.fast.pkg.FE.tw" = "collapse");
plm(form, data = data, model = "within", effect = "twoways")},
{options("plm.fast" = TRUE, "plm.fast.pkg.FE.tw" = "fixest");
plm(form, data = data, model = "within", effect = "twoways")},
{options("plm.fast" = TRUE, "plm.fast.pkg.FE.tw" = "lfe");
plm(form, data = data, model = "within", effect = "twoways")},
times = times)
summary(twowayFE, unit = "relative")
}
onewayRE <- microbenchmark(
{options("plm.fast" = FALSE); plm(form, data = data, model = "random")},
{options("plm.fast" = TRUE); plm(form, data = data, model = "random")},
times = times)
summary(onewayRE, unit = "relative")
twowayRE <- microbenchmark(
{options("plm.fast" = FALSE); plm(form, data = data, model = "random", effect = "twoways")},
{options("plm.fast" = TRUE); plm(form, data = data, model = "random", effect = "twoways")},
times = times)
summary(twowayRE, unit = "relative")
}
}
\keyword{manip}
\keyword{sysdata}
|
