pass function to ddply wrapped inside a function as part of that functions call

Question

I am hoping to use ddply within a function to summarise groups based on a user determined summary statistic (e.g. the mean, median, min, max), by passing the name of the summary function to apply as a variable in the function call. However, I'm not sure how to pass this to ddply. Simple e.g. ``` library(plyr) test.df<-data.frame(group=c("a","a","b","b"),value=c(1,5,5,15)) ddply(test.df,.(group),summarise, mean=mean(value, na.rm=TRUE)) ``` how could I set this up something like below, with the relevant function passed to ddply (additionally within a function of course, although this should be straightforward once the first problem is solved). Note each summary measure (mean etc.), will require na.rm=TRUE. I could do this by writing my own replacement function for each summary statistic, but this seems overly complex. Desired: ``` #fn<-"mean" #ddply(test.df,.(group),summarise, fn=fn(value, na.rm=TRUE)) ``` Thanks for any help people can provide. EDIT! Thanks all for these responses. I i…

Accepted Answer

The function that you provided in the question looks like it should work. (And indeed it took me a few moment to remember why it wouldn't). Here it is again, slightly rewritten for clarity (Iwastemptedtoansweryourquestionwithoutanyspacesiniteither;)

```
df <- data.frame(
  group = c("a", "a" ,"b" ,"b" ), 
  value = c(1, 5, 5, 15)
)

my_fun <- function(df, fn = "mean") {
  fn <- match.fun(fn)
  ddply(df, .(group), summarise, summary = fn(value, na.rm = TRUE))
}
```

The reason it doesn't work is a little subtle but comes down to how scoping (the process of looking up the values of variables from their names) works. `summarise()` uses non-standard evaluation to look up values in data frame, and the environment from which it was called. That works for `value`, but not for `fn` because it's not present where `summarise()` is called, i.e. in `ddply()`.

There are two solutions:

Use the `here()` function which was added to plyr to work around this problem

```
my_fun <- function(df, fn = "mean") {
  fn <- match.fun(fn)
  ddply(df, .(group), here(summarise), summary = fn(value, na.rm = TRUE))
}
my_fun(df, "mean")
```

Be slightly less concise and use an explicit function:

```
my_fun <- function(df, fn = "mean") {
  fn <- match.fun(fn)
  ddply(df, .(group), function(df) {
    summarise(df, summary = fn(value, na.rm = TRUE))
  })
}
my_fun(df, "mean")
```

I now understand how I could have avoided this problem in the first place in the design of plyr, but it requires some custom C/C++ code. It's fixed in dplyr but is unlikely to be ported back to plyr because it might break existing code.

pass function to ddply wrapped inside a function as part of that functions call

Solution

Problem