Custom aggregation on two columns naming fruit

Question

I want to aggregate two columns of a data frame by name, in the following somewhat special way:

• drop the parts column in the result by specially aggregating the two columns fruits and parts
• while the parts values for Apple, Banana and Strawberry doesn't matter and everything gets summarized, the parts values of Grape and Kiwi should become the new fruits name
• result (at bottom) should have 8 aggregated rows instead of 20

This may sound dead simple on the first sight, but after hours of trial and error I didn't find any useful solution. Here's the example:

theDF <- data.frame(dates = as.Date(c(today()+20)),
    fruits = c("Apple","Apple","Apple","Apple","Banana","Banana","Banana","Banana",
      "Strawberry","Strawberry","Strawberry","Strawberry","Grape", "Grape",
      "Grape","Grape", "Kiwi","Kiwi","Kiwi","Kiwi"),
    parts = c("Big Green Apple","Apple2","Blue Apple","XYZ Apple4",
      "Yellow Banana1","Small Banana","Banana3","Banana4",
      "Red Small Strawberry","Red StrawberryY","Big Strawberry",
       "StrawberryZ","Green Grape", "Blue Grape", "Blue Grape",
       "Blue Grape","Big Kiwi","Small Kiwi","Big Kiwi","Middle Kiwi"),
    stock = as.vector(sample(1:20)) )      

The current data frame:

The desired output:

Answer 1

We can use data.table. If there are patterns like the end character is capital letter or a number in 'parts' column to be removed, we can use sub to do that and use as a grouping variable along with 'dates' and get the sum of the 'stock'.

library(data.table)
setDT(theDF)[,.(stock = sum(stock)) , .(dates, fruits = sub("([0-9]|[A-Z])$", "", parts))]
#        dates      fruits stock
#1: 2016-06-19       Apple    46
#2: 2016-06-19      Banana    35
#3: 2016-06-19  Strawberry    38
#4: 2016-06-19 Green Grape    12
#5: 2016-06-19  Blue Grape    21
#6: 2016-06-19    Big Kiwi    37
#7: 2016-06-19  Small Kiwi    14 
#8: 2016-06-19 Middle Kiwi     7

---

Or using dplyr, we can similarly implement the same methodology.

library(dplyr)
theDF %>%
    group_by(dates, fruits = sub('([0-9]|[A-Z])$', '', parts)) %>% 
    summarise(stock = sum(stock))

Update

If there are no patterns and only based on manually identifying the elements in 'fruits', create a vector of elements, use %chin% to get the logical index in 'i', assign (:=) the values in 'parts' corresponding to the 'i' to 'fruits', then do the group by 'dates', 'fruits' and get the sum of 'stock'.

setDT(theDF)[as.character(fruits) %chin% c("Grape", "Kiwi"),
          fruits := parts][, .(stock = sum(stock)), .(dates, fruits)]

data

theDF <- structure(list(dates = structure(c(16971, 16971, 16971, 16971, 
16971, 16971, 16971, 16971, 16971, 16971, 16971, 16971, 16971, 
16971, 16971, 16971, 16971, 16971, 16971, 16971), class = "Date"), 
    fruits = structure(c(1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 5L, 
    5L, 5L, 5L, 3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L), .Label = c("Apple", 
    "Banana", "Grape", "Kiwi", "Strawberry"), class = "factor"), 
    parts = structure(c(1L, 2L, 3L, 4L, 5L, 6L, 7L, 8L, 14L, 
    15L, 16L, 16L, 11L, 10L, 10L, 10L, 9L, 13L, 9L, 12L), .Label = c("Apple1", 
    "Apple2", "Apple3", "Apple4", "Banana1", "Banana2", "Banana3", 
    "Banana4", "Big Kiwi", "Blue Grape", "Green Grape", "Middle Kiwi", 
    "Small Kiwi", "StrawberryX", "StrawberryY", "StrawberryZ"
    ), class = "factor"), stock = c(8, 19, 15, 4, 6, 18, 1, 10, 
    9, 16, 11, 2, 12, 13, 5, 3, 17, 14, 20, 7)), .Names = c("dates", 
"fruits", "parts", "stock"), row.names = c(NA, -20L), class = "data.frame")