Carolina

Carolina is a pyDAKOTA fork maintained by Equinor. Its raison d'être is to have easier building of a Python Dakota wrapper, without any MPI support. Carolina supports Python version 3.8, 3.9, 3.10, 3.11, 3.12

Installation

For Linux and MacOS:

pip install carolina

Otherwise, build Carolina youself as described below.

Building and installing Carolina

In order to build Carolina, Boost, including Boost.Python, and Dakota must be installed. This requires CMake and a C/C++ compiler. It is recommended to check the build scripts at .github/workflows/bundle_with_dakota_* where the full installation is described. The installation will likely vary across different operating systems. Roughly speaking, the following steps must be done:

1. Install CMAKE

2. Install Boost with correct python version (NOTE: you may need to edit the python version into the project-config.jam if on MacOS, see the excerpt from the MacOS install script below)

   python_version=$(python --version | sed -E 's/.*([0-9]+\.[0-9]+)\.([0-9]+).*/\1/')
   python_bin_include_lib="    using python : $python_version : $(python -c "from sysconfig import get_paths as gp; g=gp(); print(f\"$(which python) : {g['include']} : {g['stdlib']} ;\")")"
   sed -i '' "s|.*using python.*|$python_bin_include_lib|" project-config.jam
   

3. Install dakota

   • after downloading, replace <DAKOTA_VERSION> with the dakota version, for example 6.18
   • In order to install Dakota to a specific folder, use -DCMAKE_INSTALL_PREFIX="<INSTALL_DIR>" as part of the cmake invocation.

   cd dakota-<DAKOTA_VERSION>-public-src-cli
   mkdir -p build
   cd build
   cmake \
       -DCMAKE_CXX_STANDARD=14 \
       -DBUILD_SHARED_LIBS=ON \
       -DDAKOTA_PYTHON_DIRECT_INTERFACE=ON \
       -DDAKOTA_PYTHON_DIRECT_INTERFACE_NUMPY=ON \
       -DDAKOTA_DLL_API=OFF \
       -DHAVE_X_GRAPHICS=OFF \
       -DDAKOTA_ENABLE_TESTS=OFF \
       -DDAKOTA_ENABLE_TPL_TESTS=OFF \
       -DCMAKE_BUILD_TYPE='Release' \
       -DDAKOTA_NO_FIND_TRILINOS:BOOL=TRUE \
       ..
   make -j4 install
   

   This step is the one that might be the most tricky to get working on your local OS. It expects a number of packages to be found, including libgfortran, eigen, lapack, numpy, and for the appropriate libraries to be on LD_LIBRARY_PATH(linux)/DYLD_LIBRARY_PATH(MacOS). Build errors often arise from (1) the package not being installed or (2) library folders/files of the installed package not being on the library path (LD_LIBRARY_PATH for linux or DYLD_LIBRARY_PATH for MacOS).

4. After installing Dakota, it is possible to run pip install . as it will look for the following environment variables:

• The BOOST_ROOT environment variable can be set to the location of the boost library containing the folders include and lib, if they are not already included globally.

• The BOOST_PYTHON can be set if a given version of boost_python is needed. For instance if Python 3.8 is to be used:

      export BOOST_PYTHON=boost_python38
  

  By default the installation script will try to guess the boost_python version from the minor version of Python, i.e. for Python 3.10, it will try boost_python310.

• It also expects dakota binary executable to be on the system PATH. To verify this, see if you can type dakota in the terminal and run it without errors. Then, try start up python and see if you can import dakota. If these two "tests" pass, you should be able to install Carolina.

Carolina can then be installed with:

    pip install .

The library can then be tested by entering the tests directory and execute:

    pytest

In the case of testing newer versions of Dakota, scripts can be found in the script folder.

Carolina requires Dakota 6.18, but will work with older versions as well. Pathes can be reverted to allow for building against versions prior to 6.13 or 6.16.

From Dakota version 6.13 a different set of boost libraries is needed: instead of boost_signals, boost_program_options is used. From Dakota version 6.16 a small change was made in the Python interface. From Dakota version 6.18 a file was removed from the source and build script was altered.