Implement unjoin about tidyr

[hadley]

billboard <- read.csv("vignettes/billboard.csv") %>%
  tbl_df() %>%
  gather(week, rank, starts_with("wk"), na.rm = TRUE) %>%
  mutate(week = extract_numeric(week)) %>%
  arrange(artist, track, week)

song <- billboard %>% 
  select(year:time) %>%
  distinct() %>%
  mutate(song_id = 1:n())

rank <- billboard %>% 
  left_join(song) %>% 
  select(song_id, week, rank) %>% 
  distinct() %>% 

unjoin(billbord, 
  song = year:date.entered, 
  rank = c(week, rank)
)

[hadley]

I think unjoin() will need to split into two tables each time (like join combines two tables). That means you only need to specify either the x or y vars.

Here's another implementation idea:

xvars <- c("year", "artist", "track", "time", "date.entered")
id <- group_indices_(billboard, .dots = xvars)

song <- billboard[!duplicated(id), xvars]
song$sond_id <- id[!duplicated(id)]

rank <- billboard[, setdiff(names(billboard), xvars)]
rank$song_id <- id

@lionel- any thoughts?

[lionel-]

unjoin(billbord, 
  song = year:date.entered, 
  rank = c(week, rank)
)

This would yield a list of two data frames right?

Would the following work to separate group-level data?

billboard %>% unjoin(year:date.entered)

And would it be an option to have it return one data frame instead of a list in this case?

[hadley]

@lionel- I think it always has to return two tables - otherwise how can you hook them together after the fact? (i.e. unjoin() has to create a new id column)

[lionel-]

This could be dealt with an optional argument .id, e.g.:

unjoin_.data.frame <- function(.data, ..., .dots, .id) {
  dots <- lazyeval::all_dots(.dots, ...)
  if (!is.null(.id)) {
    dots[[length(dots) + 1]] <- lazyeval::as.lazy(.id)
  }
  .data <- dplyr::select_(.data, .dots = dots)

  if (is.null(.id)) {
    # Code for multiple data frames. Yields a list.

  } else {
    .data %>%
      purrr::slice_rows(.id) %>%
      purrr::by_slice(function(slice) {
        n_unique_cols <- vapply(slice, dplyr::n_distinct, numeric(1))
        if (!all(n_unique_cols == 1)) {
          stop("values are not unique within groups", call. = FALSE)
        }
      })
    res <- distinct_(.data)
  }

  res
}

Then, with a table that contains the group-level id column:

billboard_full <- billboard %>% left_join(song)
billboard_full %>% unjoin(year:time, .id = "song_id")
## Source: local data frame [317 x 5]

##    year         artist                   track time song_id
## 1  2000          2 Pac Baby Don't Cry (Keep... 4:22       1
## 2  2000        2Ge+her The Hardest Part Of ... 3:15       2
## 3  2000   3 Doors Down              Kryptonite 3:53       3
## 4  2000   3 Doors Down                   Loser 4:24       4
## 5  2000       504 Boyz           Wobble Wobble 3:35       5
## 6  2000           98^0 Give Me Just One Nig... 3:24       6
## 7  2000        A*Teens           Dancing Queen 3:44       7
## 8  2000        Aaliyah           I Don't Wanna 4:15       8
## 9  2000        Aaliyah               Try Again 4:03       9
## 10 2000 Adams, Yolanda           Open My Heart 5:30      10
## ..  ...            ...                     ...  ...     ...

I need a function that directly gives me the group-level data frame but I understand if you find this an unnecessary complication for unjoin() since it's not very difficult to subset the correct table right after separation.

[hadley]

I'm going to close this for now, since I've never actually had a use for it.