How can I avoid declaring a global variable in this R code? - r

Question

I am trying to modify Kyle Gorman's autoloess function to be callable as a method in ggplot2's stat_smooth. autoloess is a simply wrapper which runs loess through an optimiser to find the value of span which minimises AICc.
I have created something which runs successfully, but only by using a global variable. Is there a more elegant, idomatic way of programming this?
My code:
AICc.loess <- function(fit) {
# compute AIC_C for a LOESS fit, from:
#
# Hurvich, C.M., Simonoff, J.S., and Tsai, C. L. 1998. Smoothing
# parameter selection in nonparametric regression using an improved
# Akaike Information Criterion. Journal of the Royal Statistical
# Society B 60: 271–293.
#
# #param fit loess fit
# #return 'aicc' value
stopifnot(inherits(fit, 'loess'))
# parameters
n <- fit$n
trace <- fit$trace.hat
sigma2 <- sum(resid(fit) ^ 2) / (n - 1)
return(log(sigma2) + 1 + 2 * (2 * (trace + 1)) / (n - trace - 2))
}
.autoloess.magic.w <- NULL
autoloess <- function(formula, data, weights, span=c(0.01, 2.0)) {
.autoloess.magic.w <- ~weights
fit <- loess(formula=formula,
data=data,
weights=.autoloess.magic.w)
stopifnot(length(span) == 2)
# loss function in form to be used by optimize
f <- function(span) AICc.loess(update(fit, span=span))
# find best loess according to loss function
res <- update(fit, span=optimize(f, span)$minimum)
cat(paste("Optimal span:", res$pars$span, "\n"))
return(res)
}
And a quick test:
# Test
library(ggplot2)
set.seed(1984)
# Create a cubic curve
df <- data.frame(x=1:2500, y=500000 +
(-1000*(1:2500)) +
((1:2500)^2) +
-0.00025*((1:2500)^3) +
rnorm(2500, sd=60000),
ww=runif(2500, min=0, max=10))
# Use loess span
ggplot(df, aes(x=x, y=y, weight=ww)) + geom_point() + stat_smooth(method="loess")
# Use autoloess
ggplot(df, aes(x=x, y=y, weight=ww)) + geom_point() + stat_smooth(method="autoloess")

Answer 1

You can use the weight variable (seems like it is there when the function is called):
autoloess <- function(formula, data, weights, span=c(0.01, 2.0)) {
fit <- loess(formula = formula,
data = data,
weights=weight)
stopifnot(length(span) == 2)
# loss function in form to be used by optimize
f <- function(span) AICc.loess(update(fit, span=span))
# find best loess according to loss function
res <- update(fit, span=optimize(f, span)$minimum)
cat(paste("Optimal span:", res$pars$span, "\n"))
return(res)
}