Very Accurate Emulator (21cmVAE) is an emulator of the 21-cm global signal. Given an input of seven astrophyscial parameters, it directly computes realizations of the global signal across redshifts 5-50. The emulator is described in detail in Bye et. al 2022.
21cmVAE emulates global signals with an average error of 0.34% of the signal amplitude (equivalently 0.54 mK) with a run time of 40 ms on average. It is trained on ~30,000 cases, the same training set as the other exisiting emulators 21cmGEM and globalemu. The accuracy and speed makes 21cmVAE a possible tool for parameter fitting, using samplig techinques like MCMC. Moreover, the variational autoencoder approach creates an interpretable latent space that allows us to determine the relative importance of the model parameters on the global 21-cm signal.
21cmVAE is free to use on the MIT open source license. We provide here our best pretrained model, as well as code to modify and train new models. We also provide the code used for the hyperparameter tuner and code to run and train the direct emulator. We refer to the sample notebooks for an introduction on how to run and test the pretrained model, as well as how to train new models.
Questions and comments are welcome; please e-mail me at [email protected]. If you use this work for academic purposes, please cite Bye et. al 2022 and link to this repository.
Dependencies: python>=3.6, tensorflow>=2.5, h5py, jupyter, matplotlib, numpy, tqdm
For developing: black, flake8, pytest, tox
To install 21cmVAE in a virtual environment (recommended) with all required dependencies, do:
git clone https://github.com/christianhbye/21cmVAE.git
cd 21cmVAE
python -m venv .venv
source .venv/bin/activate
python -m pip install .For development, please install with python -m pip install .[dev] to get the extra dependencies.
21cmVAE uses semantic versoning.
Main author: Christian H. Bye
Suggestions and additional contributions from:
- Stephen KN Portillo
- Anastasia Fialkov
If you have suggestions for improvements/additional features, notice a bug, or want to contribute in another way, please open an issue, make a pull request or just e-mail me ([email protected]).