Combine select and mutate

问题

Quite often, I find myself manually combining select() and mutate() functions within dplyr. This is usually because I'm tidying up a dataframe, want to create new columns based on the old columns, and only want keep the new columns.

For example, if I had data about heights and widths but only wanted to use them to calculate and keep the area then I would use:

library(dplyr)
df <- data.frame(height = 1:3, width = 10:12)

df %>% 
  mutate(area = height * width) %>% 
  select(area)

When there are a lot of variables being created in the mutate step it can be difficult to make sure they're all in the select step. Is there a more elegant way to only keep the variables defined in the mutate step?

One workaround I've been using is the following:

df %>%
  mutate(id = row_number()) %>%
  group_by(id) %>%
  summarise(area = height * width) %>%
  ungroup() %>%
  select(-id)

This works but is pretty verbose, and the use of summarise() means there's a performance hit:

library(microbenchmark)

microbenchmark(

  df %>% 
    mutate(area = height * width) %>% 
    select(area),

  df %>%
    mutate(id = row_number()) %>%
    group_by(id) %>%
    summarise(area = height * width) %>%
    ungroup() %>%
    select(-id)
)

Output:

      min       lq     mean   median       uq      max neval cld
  868.822  954.053 1258.328 1147.050 1363.251 4369.544   100  a 
 1897.396 1958.754 2319.545 2247.022 2549.124 4025.050   100   b

I'm thinking there's another workaround where you can compare the original dataframe names with the new dataframe names and take the right complement, but maybe there's a better way?

I feel like I'm missing something really obvious in the dplyr documentation, so apologies if this is trivial!

回答1:

Just create your own function that combines the two steps:

mutate_only = function (.data, ...) {
    names = names(match.call(expand.dots = FALSE)$...)
    .data %>% mutate(...) %>% select(one_of(names))
}

This needs some work to function properly with standard evaluation. Unfortunately the dplyr API is currently evolving on that point so I don’t know what the recommendation for this will be in a few weeks’ time. Therefore I’ll just refer to the relevant documentation.

回答2:

Just to give more visibility to @Nate's comment, transmute() is the way to go!! From its description:

mutate() adds new variables and preserves existing; transmute() drops existing variables.

edit: to give a working example,

df %>%
  transmute(area = height * width)

is the same as

df %>% 
  mutate(area = height * width) %>% 
  select(area)

来源：https://stackoverflow.com/questions/43375156/combine-select-and-mutate

标签

dplyr