Abstract:
|
Computational pipelines for data analysis have long consisted of combinations of multiple tools. From storage backends to the extraction and agggregation layer, onto modeling layers before a reporting and graphing step: any of these components might be driven in a different programming language. At the same time, the R language and environment for statistical programing has long championed an extensible approach based on the notion of 'interfaces'. At the core of R are Fortran and C libraries, and its core C implementation always offered extensibility. More recently, the Rcpp package and its extensive use of C++ templating made the mapping of objects between R and C++ more seamless. This realizes a vision behind the S and R implementations: interactive and expressive exploration and visualization backed by efficient implementations of underlying methods. Rcpp might be the closest that we have come to realizing this vision. As of late 2015, well over 500 packages on CRAN rely on Rcpp, as are a further 70 on BioConductor. In this talk, we illustrate a few key features of Rcpp that we consider to behind the rapid adoption of Rcpp within the statistical computing community.
|