astroemu 0.1.2

Generalised framework for emulating spectral signals.

Next Generation Emulators for Cosmology and Astrophysics

Author	Harry Bevins
Version	0.1.2
Homepage	https://github.com/htjb/astroemu

UNDER DEVELOPMENT

astroemu implements a generalized framework for emulating spectral signals and is inspired by the globalemu package.

The neural network emulators are implemented in JAX and the dataloaders are built on top of PyTorch.

As with globalemu the idea is to input the independent variables alongside the physical parameters of your model then predicting a single corresponding spectral value. Full spectra can then be generated via a vectorised call to the network. The training data is tiled in the dataloaders so that the parameters and independent variables are concatenated as inputs and stacked up alongside the outputs. For example if we have a signal $y = f(x, \theta)$ and we have N $\theta$ samples and m $x$ and $y$ values then our training data looks like

Input	Output
[$\theta_{0}$, $x_0$]	$y_0$
[$\theta_{0}$, $x_1$]	$y_1$
[$\theta_{0}$, $x_2$]	$y_2$
[$\theta_{0}$, ...]	...
[$\theta_{0}$, $x_m$]	$y_m$
[..., ...]	...
[$\theta_N$, $x_m$]	$y_m$

For more details see the globalemu paper. A paper is in preparation demonstrating applications of this package to a broad range of astrophysical signals.

Contributions

Contributions are welcome! Please open an issue to discuss and have a read of the Contribution guidelines.