Basic Gaussian process regression library. (Eigen3 required)

• Multivariate Gaussian process regression
• Calculation of the derivative at a point
• Calculation of the uncertainty at a point
• Save and Load the Gaussian process to/from files
• Kernels: White, Gaussian, Periodic, RationalQuadratic, Sum and Product
• Derivative of the kernels
• Likelihood functions: Gaussian Log Likelihood (incl. derivative wrt. kernel parameter)
• Prior distributions: Gaussian, Inverse Gaussian, Gamma (incl. sampling, cdf and inverse cdf)
• Prior distributions can be built by providing their mode and variance

To setup the library clone the git repository first

git clone https://github.com/ChristophJud/GPR.git

The building of GPR is based on cmake. So navigate to the main directory GPR and create a build directory.

mkdir build	# create a build directory
cd build
ccmake ..	# ccmake is an easy tool to set config parameters

Set the build type and the installation directory and

CMAKE_CXX_FLAGS		-std=c++11

Since GPR depends on the matrix library Eigen provide its include directory

EIGEN3_INCLUDE_DIR	/path/to/eigen/eigen-3.2.4/install/include/eigen3

If not all required Boost libraries are found on the system provide a custom installation

Boost_INCLUDE_DIR 	/path/to/boost/boost_1_57_0/
Boost_LIBRARY_DIR	/path/to/boost/boost_1_57_0/stage/lib/

Make sure that boost has been built with C++11 by adding cxxflags="-std=c++11" to the b2 command.

Finally, type

make install -j8

and the library including all test programs will be built.

If you want to include the library into your own project the straight forward way is the following: Add

FIND_PACKAGE(GPR REQUIRED)

to your CMakeLists.txt file and provide

GPR_DIR			/path/to/main/gpr/project/dir/cmake 

In the CMakeLists.txt you can link your program with ${GPR_LIBRARIES}.

The tests can be seen as good examples to how to use the library.

• Matrix valued kernels
• Store/load into/from hdf5 files

A thorough introduction can be found in the open book of C.E. Rasmussen: Rasmussen, Carl Edward. Gaussian processes for machine learning. (2006).

GPR itself is licensed under the Apache 2.0 license. It depends, however, on other open source projects, which are distributed under different licenses. These are Eigen, LAPACK and Boost.