Applying group_by and summarise(sum) but keep columns with non-relevant conflicting data?

Question

My question is very similar to Applying group_by and summarise on data while keeping all the columns' info but I would like to keep columns which get excluded because th

Answer 1

Here's a tidyverse solution that keeps your group_by statement the same. The key is to use mutate_if for each variable type first (i.e., numeric, character), then get distinct rows.

library(tidyverse)
#> Loading tidyverse: ggplot2
#> Loading tidyverse: tibble
#> Loading tidyverse: tidyr
#> Loading tidyverse: readr
#> Loading tidyverse: purrr
#> Loading tidyverse: dplyr
#> Conflicts with tidy packages ----------------------------------------------
#> filter(): dplyr, stats
#> lag():    dplyr, stats

Label <- c("203c", "203c", "204a", "204a", "204a", "204a", "204a", "204a",
  "204a", "204a")
Type <- c("wholefish", "flesh", "flesh", "fleshdelip", "formula", "formuladelip",
  "formula", "formuladelip", "wholefish", "wholefishdelip")
Proportion <- c(1, 1, 0.67714, 0.67714, 0.32285, 0.32285, 0.32285, 0.32285,
  0.67714, 0.67714)
N <- (1:10)
C <- (1:10)
Code <- c("c", "a", "a", "b", "a", "b", "c", "d", "c", "d")

df <- data_frame(Label, Type, Proportion, N, C, Code)
df
#> # A tibble: 10 x 6
#>    Label           Type Proportion     N     C  Code
#>    <chr>          <chr>      <dbl> <int> <int> <chr>
#>  1  203c      wholefish    1.00000     1     1     c
#>  2  203c          flesh    1.00000     2     2     a
#>  3  204a          flesh    0.67714     3     3     a
#>  4  204a     fleshdelip    0.67714     4     4     b
#>  5  204a        formula    0.32285     5     5     a
#>  6  204a   formuladelip    0.32285     6     6     b
#>  7  204a        formula    0.32285     7     7     c
#>  8  204a   formuladelip    0.32285     8     8     d
#>  9  204a      wholefish    0.67714     9     9     c
#> 10  204a wholefishdelip    0.67714    10    10     d

df %>%
  group_by(Label, Code) %>%
  mutate_if(is.numeric, sum) %>%
  mutate_if(is.character, funs(paste(unique(.), collapse = "_"))) %>%
  distinct()
#> # A tibble: 6 x 6
#> # Groups:   Label, Code [6]
#>   Label                        Type Proportion     N     C  Code
#>   <chr>                       <chr>      <dbl> <int> <int> <chr>
#> 1  203c                   wholefish    1.00000     1     1     c
#> 2  203c                       flesh    1.00000     2     2     a
#> 3  204a               flesh_formula    0.99999     8     8     a
#> 4  204a     fleshdelip_formuladelip    0.99999    10    10     b
#> 5  204a           formula_wholefish    0.99999    16    16     c
#> 6  204a formuladelip_wholefishdelip    0.99999    18    18     d

Answer 2

Here are two other options:

1) Nest columns into one column and then customize the summary by checking the data types:

df %>% 
    group_by(Label, Code) %>% nest() %>% 
    mutate(data = map(data, 
        ~ as.tibble(map(.x, ~ if(is.numeric(.x)) sum(.x) else paste(.x, collapse="_")))
          )
    ) %>% unnest()

# A tibble: 6 x 6
#   Label   Code                        Type Proportion     N     C
#  <fctr> <fctr>                       <chr>      <dbl> <int> <int>
#1   203c      c                   wholefish    1.00000     1     1
#2   203c      a                       flesh    1.00000     2     2
#3   204a      a               flesh_formula    0.99999     8     8
#4   204a      b     fleshdelip_formuladelip    0.99999    10    10
#5   204a      c           formula_wholefish    0.99999    16    16
#6   204a      d formuladelip_wholefishdelip    0.99999    18    18

2) summarize separately and then join the result:

numeric <- df %>% 
    group_by(Label, Code) %>% 
    summarise_if(is.numeric, sum)

character <- df %>% 
    group_by(Label, Code) %>% 
    summarise_if(~ is.character(.) || is.factor(.), ~ paste(., collapse="_"))

inner_join(numeric, character, by = c("Label", "Code"))
# A tibble: 6 x 6
# Groups:   Label [?]
#   Label   Code Proportion     N     C                        Type
#  <fctr> <fctr>      <dbl> <int> <int>                       <chr>
#1   203c      a    1.00000     2     2                       flesh
#2   203c      c    1.00000     1     1                   wholefish
#3   204a      a    0.99999     8     8               flesh_formula
#4   204a      b    0.99999    10    10     fleshdelip_formuladelip
#5   204a      c    0.99999    16    16           formula_wholefish
#6   204a      d    0.99999    18    18 formuladelip_wholefishdelip

Answer 3

Here's the data.table solution, I'm assuming you want the mean() of Proportion, since these grouped proportions are likely not additive.

setDT(df)

df[, .(Type =paste(Type,collapse="_"), 
  Proportion=mean(Proportion),N= sum(N),C=sum(C)), by=.(Label,Code)]
  [order(Label)]

   Label Code                        Type Proportion  N  C
1:  203c    c                   wholefish   1.000000  1  1
2:  203c    a                       flesh   1.000000  2  2
3:  204a    a               flesh_formula   0.499995  8  8
4:  204a    b     fleshdelip_formuladelip   0.499995 10 10
5:  204a    c           formula_wholefish   0.499995 16 16
6:  204a    d formuladelip_wholefishdelip   0.499995 18 18

I'm not sure this is the cleanest dplyr solution, but it works:

df %>% group_by(Label, Code) %>% 
  mutate(Type = paste(Type,collapse="_")) %>% 
  group_by(Label,Type,Code) %>% 
  summarise(N=sum(N),C=sum(C),Proportion=mean(Proportion))

Note the key here is to re-group once you create the combined Type column.

   Label                        Type   Code     N     C Proportion
  <fctr>                       <chr> <fctr> <int> <int>      <dbl>
1   203c                       flesh      a     2     2   1.000000
2   203c                   wholefish      c     1     1   1.000000
3   204a               flesh_formula      a     8     8   0.499995
4   204a     fleshdelip_formuladelip      b    10    10   0.499995
5   204a           formula_wholefish      c    16    16   0.499995
6   204a formuladelip_wholefishdelip      d    18    18   0.499995