Tidy evaluation programming and ggplot2

Question

Trying to write a relatively simple wrapper to produce some plots, but can not work out how to specify tidy evaluation of grouping variables specified as ... an example function that facets variables but doesn't distinguish by grouping...

my_plot <- function(df = starwars,
                    select = c(height, mass),
                    ...){
    results <- list()
    ## Tidyeval arguments
    quo_select <- enquo(select)
    quo_group  <- quos(...)
    ## Filter, reshape and plot
    results$df <- df %>%
                  dplyr::filter(!is.na(!!!quo_group)) %>%
                  dplyr::select(!!quo_select, !!!quo_group) %>%
                  gather(key = variable, value = value, !!!quo_select) %>% 
                  ## Specify what to plot
                  ggplot(aes(value)) +
                  geom_histogram(stat = 'count')  +
                  facet_wrap(~variable, scales = 'free', strip.position = 'bottom')
    return(results)
}
## Plot height and mass as facets but colour histograms by hair_color
my_plot(df = starwars, select = c(height, mass), hair_color)

Great it works, but how to distinguish between different hair_color? Normally this is done within aes() but since this is using the results of quos() (i.e. quo_group) I should (I think) be using aes_() instead

my_plot <- function(df = starwars,
                    select = c(height, mass),
                    ...){
    results <- list()
    ## Tidyeval arguments
    quo_select <- enquo(select)
    quo_group  <- quos(...)
    ## Filter, reshape and plot
    results$df <- df %>%
                  dplyr::filter(!is.na(!!!quo_group)) %>%
                  dplyr::select(!!quo_select, !!!quo_group) %>%
                  gather(key = variable, value = value, !!!quo_select) %>% 
                  ## Specify what to plot, including colouring by the supplied ... groupings
                  ggplot(aes_(~value, colour = !!!quo_group)) + 
                  geom_histogram(stat = 'count')  +
                  facet_wrap(~variable, scales = 'free', strip.position = 'bottom')
    return(results)
}
## Plot height and mass as facets but colour histograms by hair_color
my_plot(df = starwars, select = c(height, mass), hair_color)
Error in !quo_group : invalid argument type

I can't see or work out having read Programming with dplyr several times now where I'm going wrong.

Can anyone point out my error/show me the way?

Answer 1

The new released ggplot2 v3.0.0 supports !! inside aes(). With some minor modification, your function is now working

library(tidyverse)

my_plot <- function(df = starwars,
                    select = c(height, mass),
                    ...){
  results <- list()

  ## Tidyeval arguments
  quo_select <- enquo(select)

  # only need quo here, if quos is used then we need to `unlist` to 
  # convert its output from list to vector
  quo_group  <- quo(...) 

  ## Filter, reshape and plot
  results$df <- df %>%
    dplyr::filter(!is.na(!!!quo_group)) %>%
    dplyr::select(!!quo_select, !!!quo_group) %>%
    gather(key = variable, value = value, !!!quo_select) %>% 
    ## Specify what to plot, including coloring by the supplied dots `...` 
    ggplot(aes(value, color = !!quo_group, fill = !!quo_group)) +  # unquote inside aes
    geom_histogram(stat = 'count')  +
    facet_wrap(vars(variable), scales = 'free', strip.position = 'bottom')
  return(results)
}

## Plot height and mass as facets but color histograms by hair_color
my_plot(df = starwars, select = c(height, mass), hair_color)

Created on 2018-09-12 by the reprex package (v0.2.0.9000).

Answer 2

I'm not sure I understand the question. Does this satisfy the requirements?

library(ggplot2)
library(data.table)

your_plot <- function(df, select, color=NULL) {

  df <- as.data.table(df)[, mget(na.omit(c(select, color)))]

  ggplot(melt(df, color, select), aes_string(x=quote(value), color=color)) +
    geom_histogram(stat="count") +
    facet_wrap(~variable, scales="free", strip.position="bottom")

}

your_plot(dplyr::starwars, c("height", "mass"), "hair_color")

This uses melt to stack the select variables, with the color variable(s) repeated for each stack. It also uses aes_string, since aes(x=value, color=color) fails when color=NULL.