Download and installation
Statistical back-end
The psgp-package includes all code necessary to apply projected squential Gaussian process (pspgp) methods in combination with the use of a population Monte Carlo algorithm for processing observations sequentially which permits the use of arbitrary sensor and noise models. The psgp-methods are advantageous in a number of scenarios. Firstly, measurement error characteristics (magnitude and distribution), if known, can be used to improve the model accuracy. Secondly, by sequentially computing a low-rank approximation to the Gaussian Process posterior dis- tribution, a fast likelihood-based estimation of the covariance structure can be carried out even for large data sets.
Installation instructions
Currently, the psgp R package is only supported on Linux. The package is written in C++ and relies on the IT++ (http://sourceforge.net/apps/wordpress/itpp/) library for linear algebra. It is thus necessary to install IT++ prior to installing the psgp package.
Installing IT++ from a repository
On Linux distributions, this can be achieved by installing the relevant packages from the repository. For example, in Ubuntu, this can be done by running the following in a terminal: $ sudo apt-get install libitpp6gf libitpp-dev libitpp6-dbg Distribution-speci?c packages are also available from the IT++ website for manual install: http://sourceforge.net/apps/wordpress/itpp/download/.
Installing IT++ from the source
Alternately, it is possible to compile the IT++ library directly from source. The source code and installation instructions are available from http://sourceforge.net/apps/wordpress/itpp/download/. Make sure that the itpp-config executable, the itpp headers and libraries are all installed in standard locations (e.g. /usr/local/bin, /usr/local/include and /usr/local/lib respectively) where the compiler will find them.
Installing the pspg package
The package is hosted on the CRAN repository. In order to install it, run from within an R session: > install.packages('psgp') This should download, compile and install the psgp R package.
Usage instructions and examples are here.
The psgp package is also part of the gptk the Gaussian process toolkit contains the psgp code, but does not include the R interface.
