I'm trying to assign a function for each element in a loop. I would like the functions use the value of the variable, but they use the last value of the variable:
assign.instrumentslist = function()
{
for(instList in lists.instruments)
{
assign(
paste("test", instList, sep="."),
function() {print(instList)},
envir = .GlobalEnv
)
}
}
lists.instruments = c("CL", "HO", "GC")
assign.instrumentslist()
test.CL()
# return "GC"
thx
Probably this is the easiest way:
assign.instrumentslist = function() {
for(instList in lists.instruments) {
local({
i <- instList
assign(
paste("test", instList, sep="."),
function() {print(i)},
pos = .GlobalEnv
)
})
}
}
The key is to create local objects (i) on the function's enclosing environment.
In this example, the environment is generated by local.
And this is a really really bad hack:
lapply(lists.instruments,
function(x) .GlobalEnv[[paste("test1", x, sep=".")]] <- function() print(x))
I can explain the behaviour but am not sure what the correct way to solve it is (environments make my head hurt).
The problem is that your anonymous function picks up an environment. it is the environment in which it will look for objects, including instList. At the first iteration, it gets this environment <environment: 0x28e19a8> and this is the current environment for your function that is being evaluated (assign.instrumentslist()):
Browse[2]> environment()
<environment: 0x28e19a8>
At the next iteration of the loop the value of instList in the current environment <environment: 0x28e19a8> gets changed to "HO". Now both test.CL() and test.HO() have the same environment so refer to the same instList, which now has value "HO". The same thing happens in the final iteration for test.GC(). The following debugging transcript shows this:
debug at #5: assign(paste("test", instList, sep = "."), function() {
print(instList)
}, envir = .GlobalEnv)
Browse[2]>
debug at #3: instList
Browse[2]> environment(test.CL)
<environment: 0x28e19a8>
Browse[2]> eval(instList, environment(test.CL))
[1] "CL"
Browse[2]>
debug at #5: assign(paste("test", instList, sep = "."), function() {
print(instList)
}, envir = .GlobalEnv)
Browse[2]>
debug at #3: instList
Browse[2]> environment(test.CL)
<environment: 0x28e19a8>
Browse[2]> eval(instList, environment(test.CL))
[1] "HO"
When the loop is finished, the evaluation environment of the assign.instrumentslist(), <environment: 0x28e19a8>, persists because it is also the environment of your three functions. They all refer to the same environment and use the value of instList that was set during the last iteration of your loop.
What's wrong with using arguments?
AsItShouldBeDone <-
function(x){
print(x)
}
> AsItShouldBeDone('CL')
[1] "CL"
That's about as much typing work: giving the CL as argument instead of an extra addition to the name of the function. And that's how you should do it.
What is wrong with your solution? :
assign(): it's very unwise to change your global environment from within a function in any language, and definitely completely against the R way of thinking. So don't. I happened to have a dataframe called test.CL in my workspace. Well, that's gone now...creating different functions with different names and all with the same code. Why is that? If you're trying to find a shortcut to assign S3 methods for different classes, try the following. In any other case, use arguments to make your function do what it should
test.CL <- test.HO <- test.GC <- function(x) print(x)
expecting that the value of a variable is hardcoded in a function from within a loop. No, it's not. R tells you this is the function :
> test.CL
function() {print(instList)}
<environment: 0x05e32224>
So what it does is quite obvious: it prints instList from the environment given in environment. Which is the environment created by your first function. Which contains the value of instList after the looping. Which is the last value.
That mechanism is hacked by the method of Koshke. It works. It's not because it works that you should use it, in contrary; hacking environments like this can have very funny side effects and is in no way stable code. Plus, as said before:
USE ARGUMENTS FOR CHRIST'S SAKE!
