DSWL package: a Python implementation of the Debiased Spatial Whittle Likelihood

The Debiased Spatial Whittle Likelihood (DSWL) package is an open-source Python package that implements the eponymous paper. The methodology allows users to efficiently infer the parameters of stationary / homogeneous spatial and spatio-temporal covariance models for univariate or multivariate processes from gridded data with potential missing observations, e.g. due to natural boundaries. It leverages the Fast Fourier Transform, and therefore can benefit from further computational gains through GPU implementations offered by PyTorch or Cupy, both made available within the package as alternative backends to Numpy. As such, DSWL on GPU allows to fit covariance models to data observed on grids with tens of millions of locations.