Code for DeepKriging on the Global Data (arXiv:2604.01689): spherical spatial prediction with DeepKriging, MRTS-sphere / Wendland bases, and universal kriging. Implementation lives under spherical_deepkriging/.
From PyPI (project page) — typical use (Python ≥ 3.10):
pip install spherical-deepkrigingOptional extras, e.g. Jupyter / plotting / everything bundled in [all]:
pip install "spherical-deepkriging[jupyter,viz]"
pip install "spherical-deepkriging[all]"From source (development) — conda-based env and repo scripts:
Needs Miniconda. On Windows, use WSL.
git clone https://github.com/STLABTW/spherical-deepkriging.git
cd spherical-deepkriging
make install-devmake install-dev creates the conda environment and installs dependencies; make build-spherical-cpp builds the MRTS-sphere C++ extension. For tests and packaging tools in a plain venv: pip install -e ".[dev]".
- Smoke test:
examples/toy/toy_sphere_deepkriging.ipynb - Colab toy notebook:
examples/toy/toy_sphere_deepkriging_colab.ipynb(also linked by the Colab badge above) - Simulations:
examples/simulation/ - Real data:
examples/real_data/
See examples/README.md for run notes.
| Basis family | Module path | Role in this project |
|---|---|---|
| MRTS-sphere | spherical_deepkriging.basis_functions.mrts_sphere |
Primary basis for spherical-coordinate modeling |
| MRTS (Euclidean) | spherical_deepkriging.basis_functions.mrts |
Euclidean reference experiments |
| Wendland | spherical_deepkriging.basis_functions.wendland |
Compact-support baseline and comparison |
@misc{huang2026deepkrigingglobaldata,
title={DeepKriging on the Global Data},
author={Hao-Yun Huang and Wen-Ting Wang and Ping-Hsun Chiang and Wei-Ying Wu},
year={2026},
eprint={2604.01689},
archivePrefix={arXiv},
primaryClass={stat.ME},
url={https://arxiv.org/abs/2604.01689},
}