- Paper: Need to run 'release' versions of trained networks. Probably on 21cmGEM train data without any effort to make the data 'uniform' as not needed to illustrate network works. Make sure to reorder xHI network inputs appropriately and fix gui'
- Sort GUI tex labels
- Tests: Some can't be run until the repo is public because of download requirements but I have tested the relevant code briefly already.
- setup.py: May need some edits
- Travis and Circle ci
- Code cov once tests are sorted
- Finish README
- Documentation
- Example Jupyter notebooks
| GlobalEmu: | Robust Global 21-cm Signal Emulation |
|---|---|
| Author: | Harry Thomas Jones Bevins |
| Version: | 1.0.0 |
| Homepage: | https://github.com/htjb/globalemu |
| Documentation: | https://globalemu.readthedocs.io/ |
You can download trained networks with the following code after pip installing or installing via the github repository:
from globalemu.downloads import download download(False).model() # Redshift-Temperature Network download(True).model() # Redshift-Neutral Fraction Network
which will produce two files in your working directory 'T_release/' and 'xHI_release/'. Each file has the respective network model in and related pre and post processing files. You can then go on to evaluate each network for a set of parameters by running:
from globalemu.eval import evaluate # [fstar, vc, fx, tau, alpha, nu_min, R_mfp] params = [1e-3, 46.5, 1e-2, 0.0775, 1.25, 1.5, 30] res = evaluate(params, base_dir='T_release/') # Redshift-Temperature Network res = evaluate(params, base_dir='xHI_release/', xHI=True) # Redshift-Neutral Fraction Network
Results are accessed via 'res.z' and 'res.signal'.
The code can also be used to train a network on your own Global 21-cm signal or neutral fraction simulations using the built in globalemu pre-processing techniques. There is some flexibility on the required astrophysical input parameters but the models are required to subscribe to the astrophysics free baseline calculation detailed in the GlobalEmu paper (see below for a reference). More details about training your own network can be found in the documentation.
The training data is available here