Merge pull request #18 from CDU-data-science-team/development

Development

R/data_and_forecast.R

@@ -1,23 +1,51 @@
 
 # data and forecasts
 
-make_tsibble <- function(data, frequency = "Daily"){
+make_tsibble <- function(data, frequency = "Daily", remove_weekends = FALSE){
 
   if(frequency == "Weekly"){
 
-    data <- data %>% 
-      dplyr::mutate(Date = lubridate::floor_date(Date, "week"),
-                    Date = tsibble::yearweek(Date))
+    return(
+      data %>% 
+        dplyr::mutate(Date = lubridate::floor_date(Date, "week"),
+                      Date = tsibble::yearweek(Date)) %>% 
+        dplyr::filter(Total_Qty >= 0) %>% 
+        dplyr::group_by(Date) %>%
+        dplyr::summarise(quantity = sum(Total_Qty, na.rm = TRUE)) %>% 
+        dplyr::ungroup() %>% 
+        head(-1) %>% # remove the last row in case it isn't a complete week
+        tsibble::tsibble(index = Date) %>% 
+        tsibble::fill_gaps(quantity = 0)
+    )
+
+  } else {
+
+    if(remove_weekends){
+
+      return(
+        data %>%
+          dplyr::filter(Total_Qty >= 0,
+                        !lubridate::wday(Date, label = TRUE) %in% c("Sat", "Sun")) %>% 
+          dplyr::group_by(Date) %>%
+          dplyr::summarise(quantity = sum(Total_Qty, na.rm = TRUE)) %>% 
+          dplyr::ungroup() %>% 
+          dplyr::mutate(observation = dplyr::row_number()) %>% 
+          tsibble::tsibble(key = observation) %>% 
+          tsibble::fill_gaps(quantity = 0)
+      )
+    } else {
+
+      return(
+        data %>%
+          dplyr::filter(Total_Qty >= 0) %>% 
+          dplyr::group_by(Date) %>%
+          dplyr::summarise(quantity = sum(Total_Qty, na.rm = TRUE)) %>% 
+          dplyr::ungroup() %>% 
+          tsibble::tsibble(index = Date) %>% 
+          tsibble::fill_gaps(quantity = 0)
+      )
+    }
   }
-
-  data %>%
-    dplyr::filter(Total_Qty >= 0) %>% 
-    dplyr::group_by(Date) %>%
-    dplyr::summarise(quantity = sum(Total_Qty, na.rm = TRUE)) %>% 
-    dplyr::ungroup() %>% 
-    head(-1) %>% # remove the last row in case it isn't a complete week
-    tsibble::tsibble(index = Date) %>% 
-    tsibble::fill_gaps(quantity = 0)
 }
 
 forecast_series <- function(data, horizon, frequency = "Daily"){

R/data_prep_functions.R

@@ -0,0 +1,45 @@
+#' Calculate mode
+#' 
+#' @description Funnily enough, R doesn't have a mode, so this is it. 
+#' Be careful because ties just return whichever is the furthest to the 
+#' beginning of the vector with no warning given
+#' 
+#' @param v Vector of numerical values
+#' @return number- mode of the supplied vector
+#' @export
+
+calc_mode <- function(v) {
+  uniqv <- unique(v)
+  uniqv[which.max(tabulate(match(v, uniqv)))]
+}
+
+#' Get holidays
+#' 
+#' @description Download English bank and other holidays from gov.uk
+#' 
+#' @return vector of dates as string YYYY-MM-DD
+#' @export
+get_holidays <- function(){
+
+  jsonlite::fromJSON(
+    "https://www.gov.uk/bank-holidays.json")$`england-and-wales`$events$date
+}
+
+#' return number of weekdays
+#' 
+#' @description return the number of weekdays between two days, excluding 
+#' weekends (obviously) and holidays (user defined)
+#' @param from Date. Date to count from (inclusive)
+#' @param to Date. Date to count to (inclusive)
+#' @param holidays. vector of dates as string YYYY-MM-DD. You can make this 
+#' (for England) with \code{\link{get_holidays}}
+#' 
+#' @return integer. Number of weekdays between two dates
+#' @export
+
+n_weekdays <- function(from, to, holidays) { 
+
+  possible_days <- seq(from, to, "days")
+  # Count all days that are not weekend and are not holidays
+  sum(!weekdays(possible_days) %in% c("Saturday", "Sunday") & !possible_days %in% holidays)
+}

tests/testthat/test-data_forecast.R

@@ -7,7 +7,7 @@ test_that("Data produced correctly", {
 
   weekly_data <- make_tsibble(test_data, frequency = "Weekly")
 
-  testthat::expect_equal(nrow(daily_data), 2312)
+  testthat::expect_equal(nrow(daily_data), 2313)
 
   testthat::expect_equal(nrow(weekly_data), 330)