xtensor-stack/xtensor

xtensor is a C++ library for working with multi-dimensional arrays. Think of it as the C++ equivalent of NumPy, the Python library widely used in data science and scientific computing. The README includes a NumPy-to-xtensor cheat sheet for users already familiar with NumPy who want to do the same operations in C++. The library is header-only, meaning you do not build it separately. You install the header files and include them in your project. It can be installed via the conda or mamba package manager, using the vcpkg dependency manager, or directly from source using CMake. xtensor supports lazy evaluation. When you write an expression like adding two arrays or computing a sine across an array, xtensor does not immediately compute the result. Instead, it builds a symbolic expression that only calculates values when you actually read them or assign the result to a concrete array. This can save memory and computation when working with large arrays. Broadcasting is also built in. Operations between arrays of different but compatible shapes follow the same rules as NumPy: a smaller array is stretched across the dimensions of the larger one to make the shapes match. Scalars and single-element dimensions broadcast to match the other array. xtensor has companion projects that let it work with arrays from other languages. xtensor-python lets it read and write NumPy arrays directly via Python's buffer protocol, without copying data. Similar bridges exist for Julia and R. Optional SIMD acceleration is available via the xsimd library, which speeds up element-wise operations on supported hardware. The library supports C++14 through C++20 depending on the version you use.