A C++ MCMC Sampler using Eigen and Boost

The aim of this project was to build an MCMC sampler for a simple problem that uses the Eigen and Boost c++ libraries.

Installation Requirements

Start by cloning this git repository.

Dependencies and Prerequisites

Eigen

Place or symlink the Eigen c++ header-only library in the include/ directory after unzipping: https://eigen.tuxfamily.org

Boost

Download boost c++ library: https://www.boost.org/.
This project primarily makes use of the header-only libraries in boost with one exception: parsing program options. To compile the program_options boost shared library,
1. Unzip the boost tarball into the include/ project directory
```
$ tar -xzvf boost_1_78_0.tar.gz
```
1. Compile the boost shared library program_options. This can be compiled into a directory that you specify with the --prefix option. Navigate to the unzipped boost directory and run the following, specifying the path to your preferred installation location in --prefix:
```
 $ ./bootstrap.sh --prefix=<path/to/lib> --with-libraries=program_options
```
Finally, create the shared boost library with
```
$ ./b2
```
- The included Makefile with this project assumes this library has been placed in ${HOME}/lib/boost_1_78_0/stage/lib, but the BOOST make variable should be changed in accordance with where this library has been created (see --prefix option above).
Note: to view other available boost libraries, you can use
```
./bootstrap.sh --show-libraries 
```

R

R is used in this project for simulating example data and plotting posteior samples, but is otherwise not needed.

See R installation instructions here: https://cran.r-project.org/.

Once R is installed, two R libraries can be installed from the R console with

> install.packages(`tidyverse` and `here`)

MCMC Sampler Program Compilation

Note: this makefile specifies g++ as the compiler and also has variables for the Eigen and Boost library locations (EIGEN and BOOST respectively). Adjust these accordingly.
Compile the MCMC sampler using the makefile by running make from the top-level of the project directory.

Example

Simulate example data from the linear model by running the R script r/simulate_data.R
Run the MCMC sampler with the compiled binary ./bin/mcmc
Plot the posterior samples with the R script r/summarize_posterior_samples.R

Note the mcmc program takes several optional arguments (run ./bin/mcmc --help for details)
Those program options can also be specified in the configuration file mcmc.conf

Allowed options:
  -h [ --help ]                         print help message
  -z [ --seed ] arg (=1)                seed for pseudorandom number generator
  -n [ --n_mcmc ] arg (=10000)          number of mcmc iterations to run
  -t [ --thin_int ] arg (=100)          mcmc thinning interval
  -x [ --x_filename ] arg (=data/X.csv) name of X data file in csv format
  -y [ --y_filename ] arg (=data/y.csv) name of y data file in csv format
  -p [ --post_beta_filename ] arg (=data/posterior_beta.csv)
                                        name of posterior beta samples data 
                                        files in csv format
  -p [ --post_sigma_sq_filename ] arg (=data/posterior_sigma_sq.csv)
                                        name of posterior sigma_sq samples data
                                        files in csv format
  --sigma_sq_beta arg (=10)             beta normal prior variance
  --a_sigma_sq arg (=0.10000000000000001)
                                        sigma_sq inverse gamma prior shape
  --b_sigma_sq arg (=1)                 sigma_sq inverse gamma prior scale

gregory-bopp / linreg-mcmc-cpp