--- title: "aweek means 'any week'" date: "`r Sys.Date()`" output: rmarkdown::html_vignette: toc: true toc_depth: 2 vignette: > %\VignetteIndexEntry{aweek means 'any week'} %\VignetteEngine{knitr::rmarkdown} %\VignetteEncoding{UTF-8} --- Introduction ============ The day in which a week starts differs depending on context. For countries like the UK, the first day of the week is the first working day, which is Monday. This definition conforms with the [ISO 8601 standard definition for the beginning of a week](https://en.wikipedia.org/wiki/ISO_week_date), but there are examples of situations where the first day of the week is different: - [The US CDC defines an "MMWR" week which starts on a Sunday](https://wwwn.cdc.gov/nndss/document/MMWR_Week_overview.pdf) - In some regions, MSF will define a week that starts on Saturday This package provides tools to convert dates to weeks and back where a week can start on any day. You can use this package for any of the following: - convert date to week starting on any day - convert week numbers and years to dates - convert week to date - convert week to week - create a factor of weeks that contains ordered levels that includes missing weeks. Converting dates to weeks ========================= You can convert dates to weeks starting on any day by using `date2week()` with the `week_start` argument. This argument can be a number from 1 to 7 representing the ISO 8601 day of the week OR it can be a string representing the day of the week in either an English locale or the locale defined on your computer. The default of this argument is the value of `get_week_start()`, which is a thin wrapper around `options("aweek.week_start", 1L)`. **Unless you have specified a default `aweek.week_start` option with `set_week_start()`, this will always be set to 1 (Monday).** > It is **highly recommended** that you set the default `aweek.week_start` either > in the beginning of your Rscript, Rmarkdown document, or in your .Rprofile. ```{r date2week} library("aweek") set_week_start("Sunday") # setting the default week_start to Sunday set.seed(2019-03-03) dat <- as.Date("2019-03-03") + sample(-6:7, 10, replace = TRUE) dat print(w <- date2week(dat)) ``` If you need a different day on the fly, you can supply an integer or character day to the `week_start` argument. ```{r date2week_week_start} # Use character days date2week(dat, week_start = "Monday") # Use ISO 8601 days date2week(dat, week_start = 1) ``` If you want to save two extra keystrokes, you can also use the `as.aweek()` method for dates, which wraps `date2week()`: ```{r as.aweek.Date} as.aweek(dat, week_start = 1) ``` What you get back is an `aweek` class object. It can be converted back to a date with either `as.Date()` or `week2date()`: ```{r date2week2date} week2date(w) as.Date(w) ``` How does it work? ----------------- The calculation of weeks from dates requires knowledge of the current day of the week and the number of days past 1 January. Week numbers are calculated in three steps: 1. Find the day of the week, relative to the week_start (d). The day of the week (d) relative to the week start (s) is calculated using the ISO week day (i) via `d = 1L + ((i + (7L - s)) %% 7L)`. 2. Find the date that represents midweek (m). The date that represents midweek is found by subtracting the day of the week (d) from 4 and adding that number of days to the current date: `m = date + (4 - d)`. 3. Find the week number (w) by counting the number of days since 1 January to (m), and use integer division by 7: `w = 1L + ((m - yyyy-01-01) %/% 7)` For example, here's how to calculate the week for Tuesday, 6 December 2016, assuming the week start is a Sunday: ```{r example_day} the_date <- as.Date("2016-12-06") jan_1 <- as.Date("2016-01-01") i <- as.POSIXlt(the_date)$wday # 2, the ISO date for Tuesday s <- 7L # week_start for sunday # 1. Find the day of the week print(d <- 1L + ((i + (7L - s)) %% 7L)) # 2. Find the date that represents midweek print(m <- the_date + (4L - d)) # 3. Find the week print(w <- 1L + as.integer(m - jan_1) %/% 7L) # Format the week sprintf("2016-W%02d-%d", w, d) ``` For the weeks around 1 January, the year is determined by the week number. If the month is January, but the week number is 52 or 53, then the year for the week (YYYY) is the calendar year (yyyy) minus 1. However, if the month is December, but the week number is 1, then the year for the week (YYYY) is the calendar year (yyyy) plus 1. The `aweek` class --------------- The result you see above is an object of class "aweek". The `aweek` class is a character that contains the `week_start` attribute. This attribute allows it to be easily converted back to a date without the user needing to enter the start day every time. You can convert a character that matches the `YYYY-Www-d` pattern to an `aweek` class object with `as.aweek()`: ```{r as.aweek.character} x <- as.aweek("2019-W10-1") x ``` > Under the hood, it checks the validity of the week string and then add the > attribute and class: > > x <- "2019-W10-1" > attr(x, "week_start") <- 7 # Sunday > class(x) <- "aweek" > If you need to remove the class, you can just use `as.character()`: ```{r ascharacter} as.character(x) ``` Best practices -------------- The `date2week()` function only checks that dates are in ISO 8601 (yyyy-mm-dd) format before converting to weeks, *and otherwise assumes that the dates are accurate* so it's strongly recommended to make sure your dates are in either `Date` or `POISXt` format and accurate before converting to weeks. The [lubridate](https://cran.r-project.org/package=lubridate) can be used for this purpose. Use `set_week_start()` at the beginning of all your scripts to explicitly define the day on which your weeks start. This can be overridden if need be in specific parts of your scripts. Otherwise, the default will be dependent on the value of `getOption("aweek.week_start", 1L)`. Because the `week_start` arguments default to `get_week_start()`, it's recommended to specify `week_start` in `date2week()` and `week2date()` if you don't have an `aweek` object. Before you combine aweek objects, confirm that they are actually aweek objects with `inherits(myObject, "aweek")`. Weekly aggregation ------------------ There are times where you would want to aggregate your days into weeks, you can do this by specifying `floor_day = TRUE` in `date2week()`. For example, here we can show the individual weeks: ```{r date2week_floor} print(wf <- date2week(dat, week_start = "Saturday", floor_day = TRUE)) table(wf) ``` If you convert this to date, then all the dates will represent the beginning of the week: ```{r date2week_floor2date} print(dwf <- week2date(wf)) weekdays(dwf) ``` If you want to aggregate your `aweek` objects after you created them, you can always use the `trunc()` function: ```{r trunc} w <- date2week(dat) w trunc(w) ``` Factors ------- Weeks can be represented as factors, which is useful for tabulations across weeks. You can use `factor = TRUE` in `date2week()` and it will automatically fill in any missing weeks. ```{r factors} dat[1] + c(0, 15) date2week(dat[1] + c(0, 15), week_start = 1, factor = TRUE) ``` If you already have an aweek object and want to convert it to a factor, you can use `factor_aweek()`: ```{r factor_aweek} factor_aweek(w) ``` Be careful when combining factors with other dates or aweek objects as they will force the other objects to be truncated as well. Weeks to weeks -------------- You can use `change_week_start()` to convert between different week definitions if you have an `aweek` object: ```{r week2week_wednesday} w # week starting on Sunday ww <- change_week_start(w, week_start = "wednesday") # same dates, starting on Wednesday ww identical(as.Date(w), as.Date(ww)) ``` ```{r week2week, R.options=list(width = 100)} # create a table with all days in the week d <- as.Date("2019-03-03") + 0:6 res <- lapply(weekdays(d), function(i) date2week(d, week_start = i)) names(res) <- weekdays(d) data.frame(res) ``` All of these columns contain the same dates: ```{r week2week2date, R.options=list(width = 100)} data.frame(lapply(res, as.Date)) ``` Combining `aweek` objects ----------------------- You can add dates, aweek objects, or characters to aweek objects: ```{r caweekdate} c(as.aweek("2010-W10-1"), res$Sunday, "2010-W12-1", as.Date(res$Monday[1]) + 14) ``` However, you *can not* combine aweek objects with different `week_start` attributes. ```{r cweek2week_err, error = TRUE} c(res$Sunday[1], res$Wednesday[2], res$Friday[3]) ``` If you want to combine different aweek objects, you must first change their `week_start` attribute: ```{r cweekweek} wed <- change_week_start(res$Wednesday, get_week_start()) fri <- change_week_start(res$Friday, get_week_start()) c(res$Sunday[1], wed[2], fri[3]) ``` ### Dates can be appended to aweek objects Dates combined with aweek objects will will be automatically converted. ```{r add_dates} c(res$Monday, as.Date("2019-04-03")) ``` ### Add characters with caution You can also add character representation of weeks, but be aware that **it is assumed that these have the same `week_start` as the first object.** ```{r add_chars} s <- c(res$Saturday, "2019-W14-3") s m <- c(res$Monday, "2019-W14-3") m ``` **These will translate into different dates** ```{r char2date} as.Date(s[7:8]) as.Date(m[7:8]) ``` Working with weeks in data frames ================================= You may encounter a situation where you have a merged data frame with weeks starting on different days. This section will cover two situations where you may have weeks as numbers and weeks as ISO-week strings. First we will create our demonstration data that represents the same week with different `week_start` attributes. ```{r date_a_frame} # create a table with all days in the week d <- as.Date("2019-03-03") + 0:6 res <- lapply(weekdays(d), function(i) date2week(d, week_start = i)) resn <- lapply(weekdays(d), function(i) date2week(d, week_start = i, numeric = TRUE)) datf <- data.frame(wday = rep(weekdays(d), each = 7), week = unlist(res), # note: unlist converts to character week_number = unlist(resn), year = 2019, stringsAsFactors = FALSE) datf$day <- substring(datf$week, 10, 11) head(datf, 10) ``` To get the weeks (numbers or strings) to aweek objects, you should use the `start` argument to specify which day of the week they start on. Internally, this translates the week to their corresponding dates and then to aweek objects with the same `week_start` attribute (which defaults to `get_week_start()`). weeks as numbers ------------ Most commonly, you will have weeks across data sets represented by numbers. These can be converted to aweek objects using the `get_aweek()` function and to dates using the `get_date()` function: ```{r get_aweek} datf$aweek <- with(datf, get_aweek(week = week_number, year = year, day = day, start = wday)) datf$date <- with(datf, get_date(week = week_number, year = year, day = day, start = wday)) head(datf, 10) ``` These functions are also useful for constructing weeks or dates on the fly if you only have a week and a year: ```{r} get_aweek(11, 2019) get_date(11, 2019) ``` ```{r remove_things, include = FALSE} datf$aweek <- NULL datf$date <- NULL ``` weeks as characters ------------------- If you have weeks formatted as ISO-week strings, then you can convert to aweek objects using `as.aweek()`: ```{r date_a_frame_2} datf$aweek <- with(datf, as.aweek(week, start = wday)) head(datf, 10) str(datf) ``` We can tabulate them to see how they transformed: ```{r tabluate_data_frame, R.options = list(width = 100)} print(with(datf, table(before = week, after = aweek)), zero.print = ".") ``` Converting weeks to dates ========================= If you receive data that contains week definitions, you can convert it back to a date if you know where the week starts. ```{r week2date} week2date("2019-W10-1", week_start = "Sunday") # 2019-03-03 week2date("2019-W10-1", week_start = "Monday") # 2019-03-04 ``` If you have an `aweek` object, however, it will use the `week_start` attribute defined in the object, even if the default `week_start` attribute is different: ```{r week2date_aweek} set_week_start("Monday") # Set the default week_start to ISO week get_week_start(w) # show the default week_start for w week2date(w) identical(week2date(w), dat) # TRUE identical(week2date(as.character(w)), dat) # FALSE ``` You can also use `as.Date()` and `as.POISXlt()` if you have an `aweek` object: ```{r asdate} as.Date(w) as.POSIXlt(w) ```