Skip to content

Training a neural network to denoise 2D 21-cm power spectra affected by cosmic variance

License

Notifications You must be signed in to change notification settings

DanielaBreitman/21cmPSDenoiser

Repository files navigation

21cmPSDenoiser - A score-based diffusion model that denoises 21-cm power spectra

21cmPSDenoiser is a package that provides a score-based diffusion model trained on 21cmFAST simulations that is capabale of significantly reducing the effect of sample variance on individual 21-cm power spectrum (PS) realisations. In Breitman+25, we find that it's better to reproduce the 21-cm PS calculation as closely as possible to the training set for optimal performance. This is especially true when applying 21cmPSDenoiser on 21-cm PS from different simulators and / or physical models. The 21-cm power spectra in the training set have been computed with tuesday, a wrapper around powerbox. In the near future, we will provide a script to calculate the 21-cm PS from a lightcone in the exact same way as was done in the training set.

The package can be installed with pip via pip install 21cmPSDenoiser and tutorials are in docs/tutorials.

If you use this code in your research, please cite Breitman+25.

About

Training a neural network to denoise 2D 21-cm power spectra affected by cosmic variance

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages