Feature Request: swap_if(), conditional swapping of values between columns

[bhive01]

Made a mistake in a partial dataset of a very manual experiment (costly to redo) and realized I needed a conditional swap between two columns.

I envision it looking something like this in practice. I wish I could figure out how to include it in a mutate(), but it requires returning two columns at once.

dt %>%
    swap_if(., condition, Column1, Column2)

My attempt at implementation leaves a lot to be desired and is hacked entirely from if_else/replace_with in dplyr.

library(tidyverse)

dt <- tibble(V1=c(1,2,4), V2=c("a","a","b"), V3=c(2,3,1))

swap_if<- function (condition, val1, val2, missing = NA) 
{
    if (!is.logical(condition)) {
        stop("`condition` must be logical", call. = FALSE)
    }
    out1 <- val1[rep(NA_integer_, length(condition))]
    out2 <- val2[rep(NA_integer_, length(condition))]

    out1[condition & !is.na(condition)] <- val2[condition & !is.na(condition)]
    out2[condition & !is.na(condition)] <- val1[condition & !is.na(condition)]

    out1[!condition & !is.na(condition)] <- val1[!condition & !is.na(condition)]
    out2[!condition & !is.na(condition)] <- val2[!condition & !is.na(condition)]

    tibble(out1, out2)
}

#swaps, but want to use in a mutate call
swap_if(dt$V2=="b", dt$V1, dt$V3)

[bhive01]

Current workaround is the following as suggested by @bpbond
Clear, but not concise. I have to swap many columns sadly so this explodes quickly.

df %>% 
    mutate(newV1=ifelse(V2=="b",V3,V1),
                 newV3=ifelse(V2=="b",V1,V3),
                 V1=newV1,
                 V3=newV3) %>%
    select(-newV1, -newV3)

[krlmlr]

@hadley: Is this a particularly common use case we want to support?

[hadley]

No, I don't think so.

However, there is a general pattern that would help resolve this issue: if summarise() and mutate() could accept data.frames/tibbles as return values, you could do this in a single step:

swap_if <- function(cond, x, y) {
  out_x <- if_else(cond, y, x)
  out_y <- if_else(!cond, x, y)

  setNames(tibble(out_x, out_y), c(expr_name(x), expr_name(y))
}

df %>% mutate(swap_if(V2 == "b", V1, V3))

I don't think the code for getting the names is quite right, but it should be something along those lines.

That would also simplify the implementation of tidy::separate(), tidyr::unite() etc because you'd be able to make use of mutate().

[krlmlr]

Returning a tibble is related with #2171.

[krlmlr]

But I was confused I think. @hadley: Wouldn't your code work with do()?

[hadley]

Yes, but do() is very inefficient.

[krlmlr]

Groupwise do(), yes. But isn't do() over the entire data frame just a single vectorized function call?

The syntax with the dot is a bit awkward, though.

[hadley]

Oh indeed, it's just that with mutate() you get the nicer implicit scoping.