Wrappers for ADOL-C in scripting languages using SWIG

R is a language and environment for statistical computing and graphics [1]. It currently is widely used in statistics and data mining. To obtain derivatives in R, one can use several non-native approaches, including the TMB system [2] and Ryacas [3]. However, none of these options support the di↵erentiation of functions expressed as R programs, as would an algorithmic di↵erentiation (AD) tool for R. Attempts to develop such a tool include radx [4]. This tool is capable of computing firstand second-order forward-mode derivatives of univariate functions. But it is no longer actively developed. Natively, inside R, the numderiv package provides methods for calculating (usually) accurate numerical first and second order derivatives [5]. Accurate calculations are done by using Richardson’s extrapolation, or, when applicable, a complex step derivative is available. A simple di↵erence method is also provided. The deriv function from the stats package computes derivatives of simple expressions, symbolically [6]. Because numerical di↵erences cannot be reliably accurate and cannot compute adjoints, there is a need to provide derivatives within R using AD tools. One method to obtaining derivatives is ADOL-C [7]. It is a mature and widely applied tool for algorithmic di↵erentiation using operator overloading in the C++ language. Because of the language dependency it can natively be used only with applications that were originally written in C or C++. Previously, a Python-wrapper [8, 9] was written for the most widely used functionality in ADOL-C. It can be used to compute first and higher order derivatives in both forward and reverse mode for applications written in Python. This wrapper was written manually, however, and must be maintained and updated manually to keep in sync with the changes and new features of the C++ library. Also only the most commonly used ADOL-C API calls were available. Given the success of manually interfacing Python with ADOL-C we have investigated an automated interfacing mechanism for ADOL-C with R and Python. We used the SWIG interface generator for this purpose. Using the interfaces that were generated, we are able to use ADOL-C from within R and Python to obtain derivatives. The rest of the document shows how SWIG was used and provides examples of ADOL-C usage.