A case study in the implementation of a Bayesian linear model using R and / or C++ libraries

As an exercise I decided to implement a Bayesian linear model using the C++ libraries Armadillo and Eigen, and then to interface the C++ code from R via Rcpp. The main purpose of this exercise was to obtain some familiarity with these libraries and packages. Additionally I am also interested in comparing the speeds of the implementations for this problem and practicing profiling techniques on the software.

A more detailed explanation of the functions / programs found in this repository can be found at Writeup/Bayes_LM_Writeup.pdf. The file also displays my findings and experiences in terms of speed comparisons and program profiling.

---

The root directory contains the following files:

1. bayes_lm_r: an R function

   • Bayes_LM.R
   • Check_Valid_Input.R

2. bayes_lm_arma: a C++-only implementation using the Armadillo library (an executable)

   • Bayes_LM_Arma.cpp
   • Parse_Args.cpp
   • Stats_Fcns_Arma.cpp

3. bayes_lm_eigen: a C++-only implementation using the Eigen library (an executable)

   • Bayes_LM_Eigen.R
   • Parse_Args.cpp
   • Stats_Fcns_Eigen.cpp

4. bayes_lm_rcpp_arma: an R function internally calling a workhorse C++ function constructed using the Armadillo library

   • Bayes_LM_Rcpp_Arma.R
   • Check_Valid_Input.R
   • Stats_Fcns_Arma.cpp

5. bayes_lm_rcpp_eigen: an R function internally calling a workhorse C++ function constructed using the Eigen library

   • Bayes_LM_Rcpp_Eigen.R
   • Check_Valid_Input.R
   • Stats_Fcns_Eigen.cpp

---

Implementation speed comparisons

Compare the speeds of the programs as the number of observations (i.e. as n) increases

Compare the speeds of the programs as the number of variables (i.e. as p) increases