[R] Optional arguments in stats::nls

[Duncan Murdoch]

On 28/02/2017 11:07 AM, Rytis Bagdziunas wrote:
> Hello everyone,
>
> I've been struggling with the usage of ellipsis argument for stats::nls
> and similar functions. In particular, nls manual indicates this:
>
> ...: Additional optional arguments.  None are used at present.

The documentation is incorrect.

> However, "none are used" seems to be slightly misleading. Here's an
> example:
>
> data <- data.frame("x" = rnorm(100),
>                    "y" = rnorm(100))
> fn <- function(y, a) a * y
>
> ## This works
> nls(y ~ fn(x, a), data = data, start = list("a" = 1))
> ## This doesn't
> nls(y ~ fn(x, a), data = data, start = list("a" = 1), myarg = FALSE)
> ## But this does
> nls(y ~ fn(x, a), data = data, start = list("a" = 1), myarg = rnorm(100))
>
> traceback() indicates that the additional argument is passed to
> model.frame.default() but doesn't appear to do anything.
>
> Is this expected behaviour?

The docs should say "...:  additional arguments that may be passed to 
model.frame()".

Likely the reason the docs are wrong is that the call to model.frame is 
put together in a tricky way, using match.call(), and it's not so 
obvious what the conditions are under which it will be called.  I don't 
see a simple fix to the docs other than what I wrote above.

Duncan Murdoch