Override a function that is imported in a namespac

问题:

As the termplot function in R is containing some weird code that is giving me annoying bugs, I want to override it in my own test code until I find a more permanent solution. Problem is that the changed function is not loaded by the mgcv package. The mgcv package loads termplot from the stats package in its namespace, using importFrom() in the NAMESPACE file.

How can I convince mgcv to use the changed termplot? I tried :

unlockBinding("termplot", as.environment("package:stats"))
assign("termplot", my.termplot, as.environment("package:stats"))
lockBinding("termplot", as.environment("package:stats"))

and when applied to lm-objects, this works and the altered termplot is used. But when using gam-objects made by the mgcv package, this doesn't work. I'm not really going to build the stats package from source if I can avoid it...

To clarify, I also tried with

assignInNamespace("termplot", my.termplot, ns="stats")
assignInNamespace("termplot", my.termplot, ns="mgcv")

in all possible combinations, before attaching mgcv, after attaching mgcv, and I didn't manage to get it working.

---

EDIT :

I tried all options given here (apart from rebuilding either package), and couldn't get it to work. The easy way out for me is using a wrapper function. That discussion can be found here. Thanks for all the tips.

---

A reproducible example :

my.termplot <- function (x) print("my new termplot")

  unlockBinding("termplot", as.environment("package:stats"))
  assignInNamespace("termplot", my.termplot, ns="stats", envir=as.environment("package:stats"))
  assign("termplot", my.termplot, as.environment("package:stats"))
  lockBinding("termplot", as.environment("package:stats"))


y <- 1:10
x <- 1:10
xx <- lm(y~x)
termplot(xx)
require(mgcv)
dat <- gamSim(1, n = 400, dist = "normal", scale = 2)
b <- gam(y ~ s(x0) + s(x1) + s(x2) + x3, data = dat)
plot(b,all=TRUE)

plot.gam calls termplot for the non-smooth terms (x3 in this case), but fails to find the new termplot function.

---

EDIT2 : apparently, my example works. I see now I solved my own question: In the first code, I didn't add both the namespace and the package in assignInNamespace. It is important to remember to change the function both in the namespace and the package before loading the other package. Thx @hadley for pointing me in the right direction, @Marek for testing the code and reporting it works, and the rest for taking the effort to answer.

回答1:

I'm stumped - I can't figure out how plot.gam is locating termplot - it's not using the ordinary scoping rules as far as I can tell. This seems to need a deeper understanding of namespaces than I currently possess.

my.termplot <- function (x) print("my new termplot")

# where is it defined?
getAnywhere("termplot")
# in package and in namespace

unlockBinding("termplot", as.environment("package:stats"))
assign("termplot", my.termplot, "package:stats")

unlockBinding("termplot", getNamespace("stats"))
assign("termplot", my.termplot, getNamespace("stats"))

getAnywhere("termplot")[1]
getAnywhere("termplot")[2]
# now changed in both places

y <- 1:10
x <- 1:10 + runif(10)
xx <- lm(y ~ x)
termplot(xx) # works

library("mgcv")
b <- gam(y ~ s(x), data = data.frame(x, y))
plot(b) # still calls the old termplot

# I'm mystified - if try and find termplot as
# seen from the environment of plot.gam, it looks 
# like what we want
get("termplot", environment(plot.gam)) 

回答2:

Try overwriting the function that you are calling termplot from. At a guess, this is plot.gam in the mgcv package.

First load the necessary package.

library(mgcv)

Here's your alternate termplot function, added to the stats namespace.

my.termplot <- function (model, ...) 
{
  message("In my.termplot")
}

unlockBinding("termplot", as.environment("package:stats"))
assign("termplot", my.termplot, as.environment("package:stats"))
lockBinding("termplot", as.environment("package:stats"))

Likewise, here's the wrapper function, added to the mgcv namespace.

my.plot.gam <- function (x, ...) 
{
  message("In my.plot.gam")
  my.termplot()
}

unlockBinding("plot.gam", as.environment("package:mgcv"))
assign("plot.gam", my.plot.gam, as.environment("package:mgcv"))
lockBinding("plot.gam", as.environment("package:mgcv"))

Here's an example to test it, taken from ?gam.

dat <- gamSim(1, n = 400, dist = "normal", scale = 2)
b <- gam(y ~ s(x0) + s(x1) + s(x2) + s(x3), data = dat)
plot(b) 

回答3:

I think the trace() function does automatically what is attempted above. Do:

trace('termplot', edit='gedit')

Where 'gedit' is the name of a text editor. The editor will open with the original code and you can paste whatever substitution code you desire.

To return to the original version just untrace('termplot')

Caveat: I tried using this when the text editor had many files open and it didn't work. So I use 'gedit', a text editor on my system that I don't use often. This way I am sure that R will open a new instance of 'gedit'.

I'm not positive this will help, but I think it's worth a try. The search sequence when there are namespaces is really confusing.