wpa/R/identify_tenure.R

120 строки
4.3 KiB
R

# --------------------------------------------------------------------------------------------
# Copyright (c) Microsoft Corporation. All rights reserved.
# Licensed under the MIT License. See LICENSE.txt in the project root for license information.
# --------------------------------------------------------------------------------------------
#' @title Tenure calculation based on different input dates, returns data summary table or histogram
#'
#' @description
#' This function calculates employee tenure based on different input dates.
#' `identify_tenure` uses the latest Date available if user selects "Date",
#' but also have flexibility to select a specific date, e.g. "1/1/2020".
#'
#' @family Data Validation
#'
#' @param data A Standard Query dataset in the form of a data frame.
#' @param end_date A string specifying the name of the date variable representing the latest date. Defaults to "Date".
#' @param beg_date A string specifying the name of the date variable representing the hire date. Defaults to "HireDate".
#' @param maxten A numeric value representing the maximum tenure.
#' If the tenure exceeds this threshold, it would be accounted for in the flag message.
#' @param return String to specify what to return.
#' Defaults to "message".
#' Other valid values include "text", "plot", "data_cleaned", "data_dirty", and "data".
#' For "data", a data frame with the `PersonId` and a calculated variable called `TenureYear` is returned.
#'
#' @examples
#' \dontrun{
#' # Add HireDate to sq_data - method #1
#' sq_data2 <- sq_data %>% mutate(HireDate = as.Date("1/1/2015", format = "%m/%d/%Y" ))
#' identify_tenure(sq_data2)
#'
#' # Add HireDate to sq_data - method #2
#' sq_data$HireDate <-
#' rep(sample(seq(as.Date('1975/01/01'),
#' as.Date('2019/11/01'), by="day"), 15119),2)
#'
#' identify_tenure(sq_data)
#' }
#'
#' @export
identify_tenure <- function(data,
end_date = "Date",
beg_date = "HireDate",
maxten = 40,
return = "message"){
required_variables <- c("HireDate")
## Error message if variables are not present
## Nothing happens if all present
data %>%
check_inputs(requirements = required_variables)
data_prep <-
data %>%
mutate(Date = as.Date(Date, format= "%m/%d/%Y"), # Re-format `Date`
end_date = as.Date(!!sym(end_date), format= "%m/%d/%Y"), # Access a symbol, not a string
beg_date = as.Date(!!sym(beg_date), format= "%m/%d/%Y")) %>% # Access a symbol, not a string
arrange(end_date) %>%
mutate(End = last(end_date))
last_date <- data_prep$End
# graphing data
tenure_summary <-
data_prep %>%
filter(Date == last_date) %>%
mutate(tenure_years = (Date - beg_date)/365) %>%
group_by(tenure_years)%>%
summarise(n = n(),.groups = 'drop')
# off person IDs
oddpeople <-
data_prep %>%
filter(Date == last_date) %>%
mutate(tenure_years = (Date - beg_date)/365) %>%
filter(tenure_years >= maxten) %>%
select(PersonId)
# message
Message <- paste0("The mean tenure is ",round(mean(tenure_summary$tenure_years,na.rm = TRUE),1)," years.\nThe max tenure is ",
round(max(tenure_summary$tenure_years,na.rm = TRUE),1),".\nThere are ",
length(tenure_summary$tenure_years[tenure_summary$tenure_years>=maxten])," employees with a tenure greater than ",maxten," years.")
if(return == "text"){
return(Message)
} else if(return == "message"){
message(Message)
} else if(return == "plot"){
suppressWarnings(
ggplot(data = tenure_summary,aes(x = as.numeric(tenure_years))) +
geom_density() +
labs(title = "Tenure - Density",
subtitle = "Calculated with `HireDate`") +
xlab("Tenure in Years") +
ylab("Density - number of employees") +
theme_wpa_basic()
)
} else if(return == "data_cleaned"){
return(data %>% filter(!(PersonId %in% oddpeople$PersonId)) %>% data.frame())
} else if(return == "data_dirty"){
return(data %>% filter((PersonId %in% oddpeople$PersonId)) %>% data.frame())
} else if(return == "data"){
data_prep %>%
filter(Date == last_date) %>%
mutate(TenureYear = as.numeric((Date - beg_date)/365)) %>%
select(PersonId, TenureYear)
} else {
stop("Error: please check inputs for `return`")
}
}