mirror of https://github.com/msberends/AMR.git
155 lines
6.5 KiB
R
Executable File
155 lines
6.5 KiB
R
Executable File
# ==================================================================== #
|
|
# TITLE #
|
|
# Antimicrobial Resistance (AMR) Data Analysis for R #
|
|
# #
|
|
# SOURCE #
|
|
# https://github.com/msberends/AMR #
|
|
# #
|
|
# LICENCE #
|
|
# (c) 2018-2021 Berends MS, Luz CF et al. #
|
|
# Developed at the University of Groningen, the Netherlands, in #
|
|
# collaboration with non-profit organisations Certe Medical #
|
|
# Diagnostics & Advice, and University Medical Center Groningen. #
|
|
# #
|
|
# This R package is free software; you can freely use and distribute #
|
|
# it for both personal and commercial purposes under the terms of the #
|
|
# GNU General Public License version 2.0 (GNU GPL-2), as published by #
|
|
# the Free Software Foundation. #
|
|
# We created this package for both routine data analysis and academic #
|
|
# research and it was publicly released in the hope that it will be #
|
|
# useful, but it comes WITHOUT ANY WARRANTY OR LIABILITY. #
|
|
# #
|
|
# Visit our website for the full manual and a complete tutorial about #
|
|
# how to conduct AMR data analysis: https://msberends.github.io/AMR/ #
|
|
# ==================================================================== #
|
|
|
|
#' Vectorised Pattern Matching with Keyboard Shortcut
|
|
#'
|
|
#' Convenient wrapper around [grepl()] to match a pattern: `x %like% pattern`. It always returns a [`logical`] vector and is always case-insensitive (use `x %like_case% pattern` for case-sensitive matching). Also, `pattern` can be as long as `x` to compare items of each index in both vectors, or they both can have the same length to iterate over all cases.
|
|
#' @inheritSection lifecycle Stable Lifecycle
|
|
#' @param x a [character] vector where matches are sought, or an object which can be coerced by [as.character()] to a [character] vector.
|
|
#' @param pattern a [character] vector containing regular expressions (or a [character] string for `fixed = TRUE`) to be matched in the given [character] vector. Coerced by [as.character()] to a [character] string if possible.
|
|
#' @param ignore.case if `FALSE`, the pattern matching is *case sensitive* and if `TRUE`, case is ignored during matching.
|
|
#' @return A [logical] vector
|
|
#' @name like
|
|
#' @rdname like
|
|
#' @export
|
|
#' @details
|
|
#' These [like()] and `%like%`/`%unlike%` functions:
|
|
#' * Are case-insensitive (use `%like_case%`/`%unlike_case%` for case-sensitive matching)
|
|
#' * Support multiple patterns
|
|
#' * Check if `pattern` is a valid regular expression and sets `fixed = TRUE` if not, to greatly improve speed (vectorised over `pattern`)
|
|
#' * Always use compatibility with Perl unless `fixed = TRUE`, to greatly improve speed
|
|
#'
|
|
#' Using RStudio? The `%like%`/`%unlike%` functions can also be directly inserted in your code from the Addins menu and can have its own keyboard shortcut like `Shift+Ctrl+L` or `Shift+Cmd+L` (see menu `Tools` > `Modify Keyboard Shortcuts...`). If you keep pressing your shortcut, the inserted text will be iterated over `%like%` -> `%unlike%` -> `%like_case%` -> `%unlike_case%`.
|
|
#' @source Idea from the [`like` function from the `data.table` package](https://github.com/Rdatatable/data.table/blob/ec1259af1bf13fc0c96a1d3f9e84d55d8106a9a4/R/like.R), although altered as explained in *Details*.
|
|
#' @seealso [grepl()]
|
|
#' @inheritSection AMR Read more on Our Website!
|
|
#' @examples
|
|
#' a <- "This is a test"
|
|
#' b <- "TEST"
|
|
#' a %like% b
|
|
#' #> TRUE
|
|
#' b %like% a
|
|
#' #> FALSE
|
|
#'
|
|
#' # also supports multiple patterns
|
|
#' a <- c("Test case", "Something different", "Yet another thing")
|
|
#' b <- c( "case", "diff", "yet")
|
|
#' a %like% b
|
|
#' #> TRUE TRUE TRUE
|
|
#' a %unlike% b
|
|
#' #> FALSE FALSE FALSE
|
|
#'
|
|
#' a[1] %like% b
|
|
#' #> TRUE FALSE FALSE
|
|
#' a %like% b[1]
|
|
#' #> TRUE FALSE FALSE
|
|
#'
|
|
#' # get isolates whose name start with 'Ent' or 'ent'
|
|
#' example_isolates[which(mo_name(example_isolates$mo) %like% "^ent"), ]
|
|
#' \donttest{
|
|
#' # faster way, only works in R 3.2 and later:
|
|
#' example_isolates[which(mo_name() %like% "^ent"), ]
|
|
#'
|
|
#' if (require("dplyr")) {
|
|
#' example_isolates %>%
|
|
#' filter(mo_name() %like% "^ent")
|
|
#' }
|
|
#' }
|
|
like <- function(x, pattern, ignore.case = TRUE) {
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
meet_criteria(pattern, allow_NA = FALSE)
|
|
meet_criteria(ignore.case, allow_class = "logical", has_length = 1)
|
|
|
|
if (all(is.na(x))) {
|
|
return(rep(FALSE, length(x)))
|
|
}
|
|
|
|
# set to fixed if no valid regex (vectorised)
|
|
fixed <- !is_valid_regex(pattern)
|
|
|
|
if (ignore.case == TRUE) {
|
|
# set here, otherwise if fixed = TRUE, this warning will be thrown: argument `ignore.case = TRUE` will be ignored
|
|
x <- tolower(x)
|
|
pattern <- tolower(pattern)
|
|
}
|
|
|
|
if (is.factor(x)) {
|
|
x <- as.character(x)
|
|
}
|
|
|
|
if (length(pattern) == 1) {
|
|
grepl(pattern, x, ignore.case = FALSE, fixed = fixed, perl = !fixed)
|
|
} else {
|
|
if (length(x) == 1) {
|
|
x <- rep(x, length(pattern))
|
|
} else if (length(pattern) != length(x)) {
|
|
stop_("arguments `x` and `pattern` must be of same length, or either one must be 1 ",
|
|
"(`x` has length ", length(x), " and `pattern` has length ", length(pattern), ")")
|
|
}
|
|
unlist(
|
|
mapply(FUN = grepl,
|
|
x = x,
|
|
pattern = pattern,
|
|
fixed = fixed,
|
|
perl = !fixed,
|
|
MoreArgs = list(ignore.case = FALSE),
|
|
SIMPLIFY = FALSE,
|
|
USE.NAMES = FALSE)
|
|
)
|
|
}
|
|
}
|
|
|
|
#' @rdname like
|
|
#' @export
|
|
"%like%" <- function(x, pattern) {
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
meet_criteria(pattern, allow_NA = FALSE)
|
|
like(x, pattern, ignore.case = TRUE)
|
|
}
|
|
|
|
#' @rdname like
|
|
#' @export
|
|
"%unlike%" <- function(x, pattern) {
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
meet_criteria(pattern, allow_NA = FALSE)
|
|
!like(x, pattern, ignore.case = TRUE)
|
|
}
|
|
|
|
#' @rdname like
|
|
#' @export
|
|
"%like_case%" <- function(x, pattern) {
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
meet_criteria(pattern, allow_NA = FALSE)
|
|
like(x, pattern, ignore.case = FALSE)
|
|
}
|
|
|
|
#' @rdname like
|
|
#' @export
|
|
"%unlike_case%" <- function(x, pattern) {
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
meet_criteria(pattern, allow_NA = FALSE)
|
|
!like(x, pattern, ignore.case = FALSE)
|
|
}
|