[Rcpp-devel] Evaluating Formula's in Rcpp

[Gabriel Hoffman]

Hi,
     I any trying to develop an R function for running 1000's of 
regressions very fast.  I will omit the technical reasons for this, but 
I would like to write code to perform the following:

for(j in 1:ncol(X) ){
     fit = myRegression( y ~ age:X[,j] )
}

This uses R's convenient 'formula' functionality to evaluate the 
interaction term in the regression.

The issue is that 'myRegression' is very complicated, high overhead, and 
takes over arguments which I have omitted for simplicity. Therefore, I 
would like to pass the formula "y ~ age:X[,j]" into a Rcpp function, and 
construct the relevant matrices in C++ using Rcpp::Environment, and 
Rcpp::Language, where I change the value of j each time.  Because, this 
would require only one entry into my C++ code, I would not have to incur 
the overhead each time.  I would like to run my analysis with a call like:

# return p-values from fitting ncol(X) regressions
myRegressionWrapper( y ~ age:X[,j], data=X)

or something like this.

Essentially I would to have the nice functionality of lm() in Rcpp to 
evaluate:

mf <- match.call(expand.dots = FALSE)
m <- match(c("formula", "data", "subset", "weights", "na.action", 
"offset"), names(mf), 0L)
mf <- mf[c(1L, m)]
mf$drop.unused.levels <- TRUE
mf[[1L]] <- as.name("model.frame")
mf <- eval(mf, parent.frame())
y <- model.response(mf, "numeric")
mt <- attr(mf, "terms")
X <- model.matrix(mt, mf, contrasts)

so I can run my custom regression function on y and X quickly for each 
value of j.

Do you know how to implement the this functionality in Rcpp or through 
some other method?

Thanks,

Gabriel Hoffman, PhD
Biological Statistics and Computational Biology
Cornell University