A purrr-less way of doing this?

[TAS]

I'm wondering what would be a "purrr-less" but still tidyverse approach to doing the following:

library(tidyverse)

df <- tribble(
  ~var1, ~var2,
  'a', 1,
  'a', 2,
  'a', 3,
  'b', 1,
  'b', 1,
  'b', 3
)

df %>% 
  group_by(var1) %>%
  mutate(
      n_lower = map_dbl(var2, ~sum(var2 < .x))
  )

My feeling is that it should be possible to do something similar to:

df %>% 
  group_by(var1) %>%
  mutate(
      n_lower = sum(var1 < var1)
  )

... by telling R in some way that the first var1 refers to the whole group of observations rather than the single one.

Any ideas?

[francisbarton]

Hey there. It's an interesting question. I'm not sure what your motivation is for not using {purrr::map_*} - which seems the ideal tool for the job as you want to iterate a function over a vector. But anyway, I had a crack at doing it with {dplyr} but my skills were not adequate despite some frustrating hours of trying!

Here are a few thoughts:

• I suspect in your original map_dbl example, if I understand you right, what you really want to say is:

df %>% 
  group_by(var1) %>%
  mutate(
    n_lower = map_dbl(var2, ~sum(var2[var2 < .x]))
  )

... this gives a different result to your example. And in your second, non-purrr example, you've written var1 when presumably you mean var2.

---

• And then I thought: you want a vector result from mutate of length 6, to fill your $n_lower column. If you do group_by then it will summarise and you'll get a result of length 2 (one sum for each group in var1). So what you really want is dplyr::filter, not group_by.

• And the tool to meet your need of distinguishing var1 as a column name from var1 as a variable argument to sum is probably tidyeval.

---

Here's my attempt, which doesn't work, however. It seems to take just the first result from row 1, which is zero, and uses it to fill every row of $n_lower. Rather than doing the mutate separately for each row of the tibble. (This is why I love {purrr}!)
I would be grateful of enlightenment on this problem myself!

lower_sum <- function(df, var1, var2) {
    df %>% 
    filter({{var1}} == var1) %>% 
    summarise(sum({{var2}}[{{var2}} < var2])) %>% 
    pull(.)
}

df %>% 
  mutate(n_lower = lower_sum(df, var1, var2))

I think I know why this doesn't work (lower_sum(df, var1, var2) just returns a single value 0 rather than a vector) but my brain hurts too much now to fix it.

I reckon this is do-able using extra/duplicate columns but that feels like cheating.

This might be useful, or this or this or this ... I'm not sure.

[nirgrahamuk]

df %>% group_by(var1) %>%
  mutate(var2_c = list(var2)) %>% 
   rowwise() %>%
  mutate(n_lower = sum(var2>var2_c))

[TAS]

Thanks both!

@ francisbarton There's no reason why I would want to not use purrr, was just wondering if/how it could be done without.

@ nirgrahamuk Interesting, thanks!

[francisbarton]

This doesn't do what I want either:

lower_sum <- function(df, var1, var2) {
    df %>% 
    filter({{var1}} == var1) %>% 
    filter({{var2}} < var2) %>% 
    add_tally(wt = {{var2}}, name = "n_lower")
}

lower_sum(df, var1, var2)