I'm struggling to understand all of the nuances when passing in something to a function in R. The rlang package and how to use it is confusing to me. Can't find a good guide on when to use all the various rlang::sym or related functions.
Anyways, I'm trying to make a function that allows the passing in of a user defined function and the related parameters. For example means, quantiles, etc. I want the user_metric to always be in quotes and also it needs to be able to contain the various arguments itself such as na.rm = TRUE & so forth. Can someone show me how this might work so whether I want to pass in 'mean' or 'mean(. , na.rm=TRUE)' both would work?
library(tidyverse)
group_by_metrics=function(data, group_col, user_metric){
metrics = data %>% group_by(!!rlang:sym(group_col)) %>% summarise_all(.funs = funs(!!rlang::syms(user_metric))
return(metrics)
}
group_by_metrics(data=mtcars, group_col='vs', user_metric='mean')
group_by_metrics(data=mtcars, group_col='vs', user_metric='mean(., na.rm = TRUE)'
group_by_metrics(data=mtcars, group_col='vs', user_metric ='quantile(., probs=0.95, na.rm = TRUE')
You have to distinguish between your first case, where you simply provide the function name, from your other cases, where you're effectively defining a lambda function. For the former, you can use match.fun to find the function by name. For the latter, convert your strings to formulas, then use purrr::as_mapper() to make functions out of them. Use ensym instead of sym to allow for unquoted arguments.
group_by_metrics <- function(.data, group_col, user_metric)
{
f <- purrr::possibly( match.fun, NULL )(user_metric)
if( is.null(f) )
f <- str_c( "~", user_metric ) %>% as.formula %>% as_mapper
.data %>% group_by(!!rlang::ensym(group_col)) %>% summarize_all( f )
}
group_by_metrics( mtcars, "vs", "quantile(., probs=0.95, na.rm=TRUE)" )
# vs mpg cyl disp hp drat wt qsec am gear carb
# <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
# 1 0 21.7 8 462. 275. 4.25 5.36 18.0 1 5 6.30
# 2 1 32.9 6 237. 123 4.47 3.45 21.2 1 4.35 4
## Using ensym instead of sym allows you to drop " for group_col
group_by_metrics( mtcars, vs, "mean" )
# vs mpg cyl disp hp drat wt qsec am gear carb
# <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
# 1 0 16.6 7.44 307. 190. 3.39 3.69 16.7 0.333 3.56 3.61
# 2 1 24.6 4.57 132. 91.4 3.86 2.61 19.3 0.5 3.86 1.79
Note that you can avoid all this conversion, if you pass the additional arguments separately, using ...:
group_by_metrics2 <- function(.data, group_col, user_metrics, ...)
{
.data %>% group_by(!!rlang::ensym(group_col)) %>%
summarize_all( user_metrics, ... )
}
group_by_metrics2( mtcars, "vs", "quantile", probs=0.05, na.rm=TRUE)
# # A tibble: 2 x 11
# vs mpg cyl disp hp drat wt qsec am gear carb
# <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
# 1 0 10.4 5.7 141. 107. 2.90 2.55 14.6 0 3 2
# 2 1 18.0 4 74.1 58.5 2.97 1.58 17.8 0 3 1
In the last example, the string quotes " are optional around both vs and quantile.