da-stdk 1.0.1

DA-STDK: Data-adaptive spatio-temporal distributional prediction (cluster-adaptive bases, conformal calibration)

DA-STDK

Reference code for cluster-aware conformal calibration in spatio-temporal distributional prediction.

• Cluster-adaptive spatial bases — centers and scales initialized from sampling density, so capacity follows heterogeneous observation patterns instead of a fixed grid.
• Cluster-aware conformal calibration — interval widths are calibrated within spatial clusters, with a global fallback when local samples are scarce.

Benchmarks in this repo use the KAUST spatio-temporal datasets (scenarios 2a/2b).

Architecture

Cluster-adaptive spatial basis, temporal basis, and covariates are concatenated and passed through a shared MLP trunk. Quantile heads predict multiple levels; cluster-aware CQR produces calibrated prediction intervals.

Install

Python 3.10+ and Poetry are enough for most use:

poetry install --with dev

Optional: Conda env via bash envs/conda/build_conda_env.sh then conda activate st-dadk.

pip install da-stdk   # after a PyPI release
# or locally:
pip install -e .

import da_stdk
from da_stdk.models import STDKMLP, create_model
from da_stdk.data.kaust_loader import load_kaust_csv_single

Run

Single training run

poetry run python scripts/train_default.py

KAUST benchmark (multiple scenarios / models)

make kaust
# or (train only, then analyze manually):
poetry run python scripts/run_kaust_data.py --config configs/config_default.yaml
poetry run python scripts/analyze_kaust_results.py --results_dir results/kaust_data_<timestamp>

make kaust runs all scenario×model combos and calls analyze_kaust_results.py when finished (--analyze). Use make kaust-dry to preview commands.

More scripts and flags: scripts/README.md.

Layout

Path	Contents
da_stdk/	Models, training, data I/O, conformal utils, viz
scripts/	Training and experiment drivers
configs/	YAML configs
data/	KAUST CSVs (large; not on PyPI)

Dev

make test          # pytest
make lint          # black, isort, mypy
pre-commit run --all-files

Citation

If you use this code, please cite:

Cluster-Aware Conformal Calibration for Spatio-Temporal Distributional Prediction Gooyoung Kim, Chae Young Lim, Wen-Ting Wang, Hao-Yun Huang, Wei-Ying Wu arXiv preprint (link forthcoming)

@misc{kim2026cluster,
  title        = {Cluster-Aware Conformal Calibration for Spatio-Temporal Distributional Prediction},
  author       = {Kim, Gooyoung and Lim, Chae Young and Wang, Wen-Ting and Huang, Hao-Yun and Wu, Wei-Ying},
  year         = {2026},
  note         = {arXiv preprint, forthcoming},
}

When the arXiv entry is available, add eprint, archivePrefix, and primaryClass (or url) to the BibTeX above.