KeopsLab logo

Python bindings for KeOps

We now fully document the public interface of the pykeops module, which is a NumPy or PyTorch front-end for the KeOps C++/Cuda library. This package contains four sets of instructions:

  1. The LazyTensor wrapper: our high-level interface, which provides a pythonic support for the most useful features of KeOps.

  2. The Genred module: our low-level syntax, compatible with NumPy and PyTorch.

  3. The generic_reduction functions: a math-friendly set of helper routines for the Genred module.

  4. kernel_product: a specific syntax with convenient aliases for operations in kernel spaces. This module is only compatible with PyTorch.