Skip to content

Machine-learning emulator for reionization-era kSZ science

A fast emulator for the kinetic Sunyaev-Zel'dovich power spectrum

reionemu helps turn simulation outputs into trainable datasets, emulator models, and reusable workflows for exploring reionization parameter space without rerunning expensive simulations.

Get Started Browse API

What the package covers

Simulation to dataset

Condense raw outputs, compute flat-sky power spectra, and assemble training-ready HDF5 datasets.

Training workflows

Build dataloaders, train deterministic or MC-dropout emulators, and evaluate validation performance with reusable utilities.

Search and tuning

Run Ray Tune experiments to explore architecture and optimizer choices for the deterministic four-parameter emulator.

Experiment artifacts

Save JSON manifests, configs, results, normalizers, and model checkpoints for reproducible emulator runs.

Start here

  • Getting Started covers installation, a quick check that the package imports, and a first training run.
  • API Overview maps the public API onto the pipeline, from simulation output to saved experiment artifact.

Repository layout

The reionemu/reionemu repository holds only the package and its documentation:

  • Core Package: src/reionemu/
  • Documentation Source: docs/

Citation

If reionemu contributes to work you publish, please cite both the software and the relevant paper.

Software

This entry uses the concept DOI, which always resolves to the latest release.

@software{pearce_reionemu,
  author    = {Pearce, Robert},
  title     = {{reionemu: Python package for emulating the kSZ angular power spectrum from reionization simulations}},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.21766410},
  url       = {https://doi.org/10.5281/zenodo.21766410},
}

GitHub's Cite this repository button generates BibTeX and APA from CITATION.cff automatically for the most current release.

Papers

An Uncertainty-Aware Machine Learning Emulator for the Reionisation kSZ Power Spectrum, Robert Pearce and Paul La Plante, in preparation (2026).