RcppEigenAD (version 1.0.0)

sourceCppAD: Construct a function to calculate the Jacobian of a function.

Description

Constructs an R function that wraps a call to a C++ function and compiles the C++ code used to define the function. The code must define a function having arguments of the type const ADmat& and a return type of ADmat. The ADmat type is a type definition for a dynamically sized matrix from the Eigen linear algebra library. (Guennebaud G, Jacob B and others 2010). The function can employ methods from the Eigen library, such as all of the standard operators from linear algebra, and a wide range of functions for calculating inverses, decompositions and so on. See Guennebaud G, Jacob B and others (2010) for further details.

The returned function can then be used as an argument to J or H which provide functions that apply algorithmic differentiation to calculate the Jacobian or Hessian matrices. The functions produced by J or H evaluate the partial derivatives with respect to the elements of the argument located at the position in the functions argument list that is specified by the wrt argument of sourceCppAD.

Usage

sourceCppAD(code=NULL,file=NULL,wrt=1,output="method")

Arguments

code

A character vector containing the C++ code to compile.

file

The filename of a file containing the C++ code to compile.

wrt

The position of the argument that J and H are differentiated with respect to.

output

Specifies what is generated by sourceCppAD. If output="method" (the default) then a function which invokes the compiled code is returned. If output="code" then the source code that wraps the users function for use with the algorithmic differentiation libraries is returned.

Value

A function which invokes the compiled code or the c++ code that wraps the users function for use with the algorithmic differentiation libraries.

References

eigenwebRcppEigenAD

Examples

# NOT RUN {
library(RcppEigenAD)
# define a function to calculate sin(cos(x)) where x is a matrix
f<-sourceCppAD('
ADmat f(const ADmat& X)
{
  return X.array().cos().sin();
}
')
x<-matrix(c(1,2,3,4),2,2)
# call it
f(x)
# }