How to summarize by group while retrieving values from columns that weren't summarized

Question

I'm trying to summarize a data frame, while grouping by a variable. My problem is that when doing such summarizing procedure, I lose other columns that I need.

Consider the following data:

df <- 
  tibble::tribble(
    ~id, ~year, ~my_value,
    1,   2010,  2,
    1,   2013,  2,
    1,   2014,  2,
    2,   2010,  4,
    2,   2012,  3,
    2,   2014,  4,
    2,   2015,  2,
    3,   2015,  3,
    3,   2010,  3,
    3,   2011,  3
  )

I want to group by id in order to collapse my_value to a single value. I use the following algorithm:

1. IF all values of my_value are identical, then simply return the first value, i.e, my_value[1].
2. ELSE return the smallest value, i.e., min(my_value).

So I wrote a small function that does it:

my_func <- function(x) {
  if (var(x) == 0) {
    return(x[1])
  }
  # else:
  min(x)
}

And now I can use either dplyr or data.table to summarize by id:

library(dplyr)
library(data.table)

# dplyr
df %>%
  group_by(id) %>%
  summarise(my_min_val = my_func(my_value))
#> # A tibble: 3 x 2
#>      id my_min_val
#>   <dbl>      <dbl>
#> 1     1          2
#> 2     2          2
#> 3     3          3

# data.table
setDT(df)[, .(my_min_val = my_func(my_value)), by = "id"]
#>    id my_min_val
#> 1:  1          2
#> 2:  2          2
#> 3:  3          3

So far so good. My problem is that I lost the year value. I want the respective year value for each chosen my_value.

My desired output should look like:

# desired output
desired_output <- 
  tribble(~id, ~my_min_val, ~year,
          1,   2,           2010,  # because for id 1, var(my_value) is 0, and hence my_value[1] corresponds to year 2010
          2,   2,           2015,  # because for id 2, var(my_value) is not 0, and hence min(my_value) (which is 2) corresponds to year 2015
          3,   3,           2015)  # because for id 3, var(my_value) is 0, hence my_value[1] corresponds to year 2015

I especially seek a data.table solution because my real data is very large (over 1 million rows) and with many groups. Thus efficiency is important. Thanks!

Answer 1

We may use the condition in slice

library(dplyr)
my_func <- function(x) if(var(x) == 0) 1 else which.min(x)
df %>% 
   group_by(id) %>% 
   slice(my_func(my_value)) %>%
   ungroup

-output

# A tibble: 3 × 3
     id  year my_value
  <dbl> <dbl>    <dbl>
1     1  2010        2
2     2  2015        2
3     3  2015        3

---

Or using data.table

library(data.table)
setDT(df)[df[, .I[my_func(my_value)], id]$V1]
   id year my_value
1:  1 2010        2
2:  2 2015        2
3:  3 2015        3

---

Or with slice_min and with_ties = FALSE

df %>%
    group_by(id) %>% 
    slice_min(n = 1, order_by = my_value, with_ties = FALSE)  %>%
    ungroup

-output

# A tibble: 3 × 3
     id  year my_value
  <dbl> <dbl>    <dbl>
1     1  2010        2
2     2  2015        2
3     3  2015        3