nomissbowling/ndarray

NumPy-compatible multidimensional arrays in C++

ndarray: NumPy-friendly multidimensional arrays in C++

ndarray is a template library that provides multidimensional array objects in C++, with an interface and features designed to mimic the Python 'numpy' package as much as possible.

More information can be found in the documentation at ndarray.github.io/ndarray.

Building from Git

CMake is the supported way to build ndarray.

To build with cmake, do:

mkdir build
cd build
cmake ..
make
make test

Inclusion and testing of optional dependencies is controlled by NDARRAY_* cmake options. Dependency resolution can be controlled by the PYBIND11_DIR, EIGEN_DIR, and FFTW_DIR environment variables. For example, to build with an alternate Eigen3 install location and disable FFTW testing replace cmake .. with EIGEN_DIR=/opt/local cmake -DNDARRY_FFTW=OFF ...

ndarray::EigenView is not compatible with Eigen >= 3.3, nor is it compatible with pybind11's default Eigen wrappers. To build without ndarray::EigenView use cmake option -DNDARRAY_EIGENVIEW=OFF

ndarray's build system does not produce the correct suffixes for pybind11 outputs under pybind11 2.1.x (due to a bug in pybind11 itself). To avoid this problem, please upgrade to pybind11 2.2.x, or try the (now reverted) patch from ndarray commit f46c0f0ff876ceab5aaa3286e5f6e86902e72feb.

However, the cmake build does not support Boost.NumPy library and Boost.Python wrappers, which are deprecated in favor of pybind11. If you are keen to use Boost.Python you can try the older scons build files, as follows

ndarray includes the Boost.NumPy library using git's "submodules" feature. When you clone the ndarray repository with git, you'll get an empty Boost.NumPy directory. Even if you don't plan to use Boost.NumPy (which is required only if you want to build the Boost.Python bindings for ndarray), it is necessary to checkout the Boost.NumPy source (as parts of the build system is shared). So, immediately after cloning ndarray, you'll need to run:

git submodule update --init --recursive

From there, you'll be able to build ndarray and (optionally) Boost.NumPy together just by running "scons" from the root of the ndarray clone.

Version 1.4.2 of ndarray is the last version to support SWIG.

Building from Compressed Source

GitHub's automatically generated tarballs and zip files don't include the Boost.NumPy submodule or the git metadata needed to run "git submodule", so these features can't be used from release tarball. Please download from git if you want to use the Boost.NumPy interface.