mirror of
https://github.com/agdamsbo/REDCapCAST.git
synced 2024-11-24 22:11:55 +01:00
314 lines
9.0 KiB
R
314 lines
9.0 KiB
R
utils::globalVariables(c("calculations", "choices"))
|
|
#' Doc table to data dictionary - EARLY, DOCS MISSING
|
|
#'
|
|
#' @description
|
|
#' Works well with `project.aid::docx2list()`.
|
|
#' Allows defining a database in a text document (see provided template) for
|
|
#' an easier to use data base creation. This approach allows easier
|
|
#' collaboration when defining the database. The generic case is a data frame
|
|
#' with variable names as values in a column. This is a format like the REDCap
|
|
#' data dictionary, but gives a few options for formatting.
|
|
#'
|
|
#' @param data tibble or data.frame with all variable names in one column
|
|
#' @param instrument.name character vector length one. Instrument name.
|
|
#' @param col.variables variable names column (default = 1), allows dplyr
|
|
#' subsetting
|
|
#' @param list.datetime.format formatting for date/time detection.
|
|
#' See `case_match_regex_list()`
|
|
#' @param col.description descriptions column, allows dplyr
|
|
#' subsetting. If empty, variable names will be used.
|
|
#' @param col.condition conditions for branching column, allows dplyr
|
|
#' subsetting. See `char2cond()`.
|
|
#' @param col.subheader sub-header column, allows dplyr subsetting.
|
|
#' See `format_subheader()`.
|
|
#' @param subheader.tag formatting tag. Default is "h2"
|
|
#' @param condition.minor.sep condition split minor. See `char2cond()`.
|
|
#' Default is ",".
|
|
#' @param condition.major.sep condition split major. See `char2cond()`.
|
|
#' Default is ";".
|
|
#' @param col.calculation calculations column. Has to be written exact.
|
|
#' Character vector.
|
|
#' @param col.choices choices column. See `char2choice()`.
|
|
#' @param choices.char.sep choices split. See `char2choice()`. Default is "/".
|
|
#' @param missing.default value for missing fields. Default is NA.
|
|
#'
|
|
#' @return tibble or data.frame (same as data)
|
|
#' @export
|
|
#'
|
|
#' @examples
|
|
#' # data <- dd_inst
|
|
#' # data |> doc2dd(instrument.name = "evt",
|
|
#' # col.description = 3,
|
|
#' # col.condition = 4,
|
|
#' # col.subheader = 2,
|
|
#' # col.calculation = 5,
|
|
#' # col.choices = 6)
|
|
doc2dd <- function(data,
|
|
instrument.name,
|
|
col.variables = 1,
|
|
list.datetime.format = list(
|
|
date_dmy = "_dat[eo]$",
|
|
time_hh_mm_ss = "_ti[md]e?$"
|
|
),
|
|
col.description = NULL,
|
|
col.condition = NULL,
|
|
col.subheader = NULL,
|
|
subheader.tag = "h2",
|
|
condition.minor.sep = ",",
|
|
condition.major.sep = ";",
|
|
col.calculation = NULL,
|
|
col.choices = NULL,
|
|
choices.char.sep = "/",
|
|
missing.default = NA) {
|
|
data <- data |>
|
|
dplyr::mutate(dplyr::across(dplyr::everything(), ~ dplyr::na_if(.x, c(""))))
|
|
|
|
|
|
## Defining the field name
|
|
out <- data |>
|
|
dplyr::mutate(
|
|
field_name = dplyr::pick(col.variables) |> unlist()
|
|
)
|
|
|
|
## Defining the field label. Field name is used if no label is provided.
|
|
if (is_missing(col.description)) {
|
|
out <- out |>
|
|
dplyr::mutate(
|
|
field_label = field_name
|
|
)
|
|
} else {
|
|
out <- out |>
|
|
dplyr::mutate(
|
|
field_label = dplyr::pick(col.description) |> unlist()
|
|
)
|
|
}
|
|
|
|
## Defining the sub-header
|
|
if (!is_missing(col.subheader)) {
|
|
out <- out |>
|
|
dplyr::mutate(
|
|
section_header = dplyr::pick(col.subheader) |>
|
|
unlist() |>
|
|
format_subheader(tag = subheader.tag)
|
|
)
|
|
}
|
|
|
|
## Defining the choices
|
|
if (is_missing(col.choices)) {
|
|
out <- out |>
|
|
dplyr::mutate(
|
|
choices = missing.default
|
|
)
|
|
} else {
|
|
out <- out |>
|
|
dplyr::mutate(
|
|
choices = dplyr::pick(col.choices) |>
|
|
unlist() |>
|
|
char2choice(char.split = choices.char.sep)
|
|
)
|
|
}
|
|
|
|
|
|
|
|
|
|
## Defining the calculations
|
|
if (is_missing(col.calculation)) {
|
|
out <- out |>
|
|
dplyr::mutate(
|
|
calculations = missing.default
|
|
)
|
|
} else {
|
|
# With inspiration from textclean package, curly apostrophe is replaced
|
|
out <- out |>
|
|
dplyr::mutate(
|
|
calculations = dplyr::pick(col.calculation) |>
|
|
unlist() |>
|
|
tolower() |>
|
|
replace_curly_quote()
|
|
)
|
|
}
|
|
|
|
## Merging choices and calculations, defining field type and setting form name
|
|
out <- out |>
|
|
dplyr::mutate(
|
|
select_choices_or_calculations = dplyr::coalesce(calculations, choices),
|
|
field_type = dplyr::case_when(!is.na(choices) ~ "radio",
|
|
!is.na(calculations) ~ "calc",
|
|
.default = "text"
|
|
),
|
|
form_name = instrument.name
|
|
)
|
|
|
|
## Defining branching logic from conditions
|
|
if (is_missing(col.condition)) {
|
|
out <- out |>
|
|
dplyr::mutate(
|
|
branching_logic = missing.default
|
|
)
|
|
} else {
|
|
out <- out |>
|
|
dplyr::mutate(
|
|
branching_logic = dplyr::pick(col.condition) |>
|
|
unlist() |>
|
|
char2cond(minor.split = condition.minor.sep,
|
|
major.split = condition.major.sep)
|
|
)
|
|
}
|
|
|
|
## Detecting data/time formatting from systematic field names
|
|
if (is.null(list.datetime.format)) {
|
|
out <- out |>
|
|
dplyr::mutate(
|
|
text_validation_type_or_show_slider_number = missing.default
|
|
)
|
|
} else {
|
|
out <- out |>
|
|
dplyr::mutate(
|
|
text_validation_type_or_show_slider_number = case_match_regex_list(
|
|
field_name,
|
|
list.datetime.format
|
|
)
|
|
)
|
|
}
|
|
|
|
## Selecting relevant columns
|
|
out <- out |>
|
|
dplyr::select(dplyr::any_of(names(REDCapCAST::redcapcast_meta)))
|
|
|
|
## Merging and ordering columns for upload
|
|
out |>
|
|
list(REDCapCAST::redcapcast_meta |> dplyr::slice(0)) |>
|
|
dplyr::bind_rows() |>
|
|
dplyr::select(names(REDCapCAST::redcapcast_meta))
|
|
}
|
|
|
|
|
|
|
|
|
|
#' Simple function to generate REDCap choices from character vector
|
|
#'
|
|
#' @param data vector
|
|
#' @param char.split splitting character(s)
|
|
#' @param raw specific values. Can be used for options of same length.
|
|
#' @param .default default value for missing. Default is NA.
|
|
#'
|
|
#' @return vector
|
|
#' @export
|
|
#'
|
|
#' @examples
|
|
#' char2choice(c("yes/no"," yep. / nope ","",NA,"what"),.default=NA)
|
|
char2choice <- function(data, char.split = "/", raw = NULL,.default=NA) {
|
|
ls <- strsplit(x = data, split = char.split)
|
|
|
|
ls |>
|
|
purrr::map(function(.x) {
|
|
if (is.null(raw)) {
|
|
raw <- seq_len(length(.x))
|
|
}
|
|
if (length(.x) == 0 | all(is.na(.x))) {
|
|
.default
|
|
} else {
|
|
paste(paste0(raw, ", ",trimws(.x)), collapse = " | ")
|
|
}
|
|
}) |>
|
|
purrr::list_c()
|
|
}
|
|
|
|
#' Simple function to generate REDCap branching logic from character vector
|
|
#'
|
|
#' @param data vector
|
|
#' @param .default default value for missing. Default is NA.
|
|
#' @param minor.split minor split
|
|
#' @param major.split major split
|
|
#' @param major.sep argument separation. Default is " or ".
|
|
#'
|
|
#' @return vector
|
|
#' @export
|
|
#'
|
|
#' @examples
|
|
#' #data <- dd_inst$betingelse
|
|
#' #c("Extubation_novent, 2; Pacu_delay, 1") |> char2cond()
|
|
char2cond <- function(data, minor.split = ",", major.split = ";", major.sep = " or ", .default = NA) {
|
|
strsplit(x = data, split = major.split) |>
|
|
purrr::map(function(.y) {
|
|
strsplit(x = .y, split = minor.split) |>
|
|
purrr::map(function(.x) {
|
|
if (length(.x) == 0 | all(is.na(.x))) {
|
|
.default
|
|
} else {
|
|
glue::glue("[{trimws(tolower(.x[1]))}]='{trimws(.x[2])}'")
|
|
}
|
|
}) |>
|
|
purrr::list_c() |>
|
|
glue::glue_collapse(sep = major.sep)
|
|
}) |>
|
|
purrr::list_c()
|
|
}
|
|
|
|
#' List-base regex case_when
|
|
#'
|
|
#' @description
|
|
#' Mimics case_when for list of regex patterns and values. Used for date/time
|
|
#' validation generation from name vector. Like case_when, the matches are in
|
|
#' order of priority.
|
|
#' Primarily used in REDCapCAST to do data type coding from systematic variable
|
|
#' naming.
|
|
#'
|
|
#' @param data vector
|
|
#' @param match.list list of case matches
|
|
#' @param .default Default value for non-matches. Default is NA.
|
|
#'
|
|
#' @return vector
|
|
#' @export
|
|
#'
|
|
#' @examples
|
|
#' case_match_regex_list(
|
|
#' c("test_date", "test_time", "test_tida", "test_tid"),
|
|
#' list(date_dmy = "_dat[eo]$", time_hh_mm_ss = "_ti[md]e?$")
|
|
#' )
|
|
case_match_regex_list <- function(data, match.list, .default = NA) {
|
|
match.list |>
|
|
purrr::imap(function(.z, .i) {
|
|
dplyr::if_else(grepl(.z, data), .i, NA)
|
|
}) |>
|
|
(\(.x){
|
|
dplyr::coalesce(!!!.x)
|
|
})() |>
|
|
(\(.x){
|
|
dplyr::if_else(is.na(.x), .default, .x)
|
|
})()
|
|
}
|
|
|
|
#' Multi missing check
|
|
#'
|
|
#' @param data character vector
|
|
#' @param nas character vector of strings considered as NA
|
|
#'
|
|
#' @return logical vector
|
|
is_missing <- function(data,nas=c("", "NA")) {
|
|
if (is.null(data)) {
|
|
TRUE
|
|
} else {
|
|
is.na(data) | data %in% nas
|
|
}
|
|
}
|
|
|
|
|
|
#' Replace curly apostrophes and quotes from word
|
|
#'
|
|
#' @description
|
|
#' Copied from textclean, which has not been updated since 2018 and is not
|
|
#' on CRAN. Github:https://github.com/trinker/textclean
|
|
#'
|
|
#' @param x character vector
|
|
#'
|
|
#' @return character vector
|
|
replace_curly_quote <- function(x){
|
|
replaces <- c('\x91', '\x92', '\x93', '\x94')
|
|
Encoding(replaces) <- "latin1"
|
|
for (i in 1:4) {
|
|
x <- gsub(replaces[i], c("'", "'", "\"", "\"")[i], x, fixed = TRUE)
|
|
}
|
|
x
|
|
}
|