stanFunction: Compile and Call a Stan Math Function

In StanHeaders: C++ Header Files for Stan

Compile and Call a Stan Math Function

Description

Call a function defined in the Stan Math Library from R using this wrapper around cppFunction.

Usage

  stanFunction(function_name, ..., env = parent.frame(), rebuild = FALSE,
               cacheDir = getOption("rcpp.cache.dir", tempdir()), 
               showOutput = verbose, verbose = getOption("verbose"))

Arguments

function_name	A character vector of length one that is the unscoped basename of a C++ function under the prim/ directory of the Stan Math Library that you would like to evaluate. Functions (such as integrate_1d) of other functions are not permitted and neither are functions (such as reject) of characters.
...	Further arguments that are passed to function_name in tag = value form, which are passed to function_name by position. See the Details and Examples sections.
env,rebuild,cacheDir,showOutput,verbose	The same as in cppFunction

Details

The stanFunction function essentially compiles and evaluates a C++ function of the form

auto function_name(...) { return stan::math::function_name(...); }

It is essential to pass all arguments to function_name through the ... in order for the C++ wrapper to know what the argument types are. The mapping between R types and Stan types is

R type	Stan type
double	real
integer	int
vector	vector
matrix(*, nrow = 1)	row_vector
matrix	matrix

and, in addition, lists of the aforementioned R types map to arrays of Stan types and thus must not be ragged if they are nested. The Stan version of the function is called with arguments specified by position, i.e. in the order that they appear in the .... However, the R wrapper function has arguments whose names are the same as the names passed through the ....

Value

The result of function_name evaluated at the arguments that are passed through the ..., which could be of various R types. It also has the side effect of defining a function named function_name in the environment given by the env argument that can subsequently be called with inputs of the same type (but not necessarily the same value) that were passed through the ....

Examples

  files <- dir(system.file("include", "stan", "math", "prim",
                           package = "StanHeaders"), 
               pattern = "hpp$", recursive = TRUE)
  functions <- sub("\\.hpp$", "", 
                   sort(unique(basename(files[dirname(files) != "."]))))
  length(functions) # you could call most of these Stan functions
  
  ## Not run: 
    log(sum(exp(exp(1)), exp(pi))) # true value
    
    stanFunction("log_sum_exp", x = exp(1), y = pi)
    args(log_sum_exp) # now exists in .GlobalEnv
    log_sum_exp(x = pi, y = exp(1))
    
    # but log_sum_exp() was not defined for a vector or matrix
    x <- c(exp(1), pi)
    try(log_sum_exp(x))
    stanFunction("log_sum_exp", x = x) # now it is
    
    # log_sum_exp() is now also defined for a matrix
    log_sum_exp(as.matrix(x))
    log_sum_exp(t(as.matrix(x)))
    log_sum_exp(rbind(x, x))
    
    # but log_sum_exp() was not defined for a list
    try(log_sum_exp(as.list(x)))
    stanFunction("log_sum_exp", x = as.list(x)) # now it is

    # in rare cases, passing a nested list is needed
    stanFunction("dims", x = list(list(1:3)))
    
    # nullary functions work but are not that interesting
    stanFunction("negative_infinity")
    
    # PRNG functions work by adding a seed argument
    stanFunction("lkj_corr_rng", K = 3L, eta = 1)
    args(lkj_corr_rng) # has a seed argument
  
## End(Not run)

StanHeaders documentation built on Sept. 8, 2023, 5:54 p.m.