R: dplyr - Group by field dynamically

‘regroup’ is deprecated / no applicable method for ‘as.lazy’ applied to an object of class “list”

A few months ago I wrote a blog explaining how to dynamically/programatically group a data frame by a field using dplyr but that approach has been deprecated in the latest version.

To recap, the original function looked like this:

library(dplyr)
 
groupBy = function(df, field) {
  df %.% regroup(list(field)) %.% summarise(n = n())
}

And if we execute that with a sample data frame we’ll see the following:

> data = data.frame(
      letter = sample(LETTERS, 50000, replace = TRUE),
      number = sample (1:10, 50000, replace = TRUE)
  )
 
> groupBy(data, 'letter') %>% head(5)
Source: local data frame [5 x 2]
 
  letter    n
1      A 1951
2      B 1903
3      C 1954
4      D 1923
5      E 1886
Warning messages:
1: %.% is deprecated. Please use %>% 
2: %.% is deprecated. Please use %>% 
3: 'regroup' is deprecated.
Use 'group_by_' instead.
See help("Deprecated")

I replaced each of the deprecated operators and ended up with this function:

groupBy = function(df, field) {
  df %>% group_by_(list(field)) %>% summarise(n = n())
}

Now if we run that:

> groupBy(data, 'letter') %>% head(5)
Error in UseMethod("as.lazy") : 
  no applicable method for 'as.lazy' applied to an object of class "list"

It turns out the ‘group_by_’ function doesn’t want to receive a list of fields so let’s remove the call to list:

groupBy = function(df, field) {
  df %>% group_by_(field) %>% summarise(n = n())
}

And now if we run that:

> groupBy(data, 'letter') %>% head(5)
Source: local data frame [5 x 2]
 
  letter    n
1      A 1951
2      B 1903
3      C 1954
4      D 1923
5      E 1886

Good times! We get the correct result and no deprecation messages.

If we want to group by multiple fields we can just pass in the field names like so:

groupBy = function(df, field1, field2) {
  df %>% group_by_(field1, field2) %>% summarise(n = n())
}

> groupBy(data, 'letter', 'number') %>% head(5)
Source: local data frame [5 x 3]
Groups: letter
 
  letter number   n
1      A      1 200
2      A      2 218
3      A      3 205
4      A      4 176
5      A      5 203

Or with this simpler version:

groupBy = function(df, ...) {
  df %>% group_by_(...) %>% summarise(n = n())
}

> groupBy(data, 'letter', 'number') %>% head(5)
Source: local data frame [5 x 3]
Groups: letter
 
  letter number   n
1      A      1 200
2      A      2 218
3      A      3 205
4      A      4 176
5      A      5 203

I realised that we can actually just use the group_by itself and pass in the field names without quotes, something I couldn’t get to work in earlier versions:

groupBy = function(df, ...) {
  df %>% group_by(...) %>% summarise(n = n())
}

> groupBy(data, letter, number) %>% head(5)
Source: local data frame [5 x 3]
Groups: letter
 
  letter number   n
1      A      1 200
2      A      2 218
3      A      3 205
4      A      4 176
5      A      5 203

We could even get a bit of pipelining going on if we fancied it:

> data %>% groupBy(letter, number) %>% head(5)
Source: local data frame [5 x 3]
Groups: letter
 
  letter number   n
1      A      1 200
2      A      2 218
3      A      3 205
4      A      4 176
5      A      5 203

And as of dplyr 0.3 we can simplify our groupBy function to make use of the new count function which combines group_by and summarise:

groupBy = function(df, ...) {
  df %>% count(...)
}

> data %>% groupBy(letter, number) %>% head(5)
Source: local data frame [5 x 3]
Groups: letter
 
  letter number   n
1      A      1 200
2      A      2 218
3      A      3 205
4      A      4 176
5      A      5 203