Welcome to SWIFT-Emulator’s Documentation
The SWIFT-Emulator is a python toolkit for using Gaussian Process machine learning to produce synthetic simulation data by interpolating between base outputs. It excels at creating synthetic scaling relations across large swathes of model parameter space, as it was created to model galaxy scaling relations as a function of galaxy formation model parameters for calibration purposes.
It includes functionality to:
Generate parameters to perform ground truth runs with in an efficient way as a latin hypercube.
Train machine learning models, including linear models and Gaussian Process Regression models (with mean models), on this data in a very clean way.
Generate densly populated synthetic data across the original parameter space, and tools to generate complex model discrepancy descriptions (known here as penalty functions).
Generate sweeps across model parameter space for the emulated scaling relations to assist in physical insight, as well as sensitivity analysis tools based upon raw and synthetic data.
Validate predictions through cross validation.
Visualise the resulting penalty data to assist in model choice decisions.
Produce inputs and read outputs from the cosmological code SWIFT that processed by VELOCIraptor and the swift-pipeline.
Information about SWIFT can be found here, Information about VELOCIraptor can be found here and tnformation about the SWIFT-pipeline can be found here.
By combining a selection of SWIFT-io and GP analysis tools, the SWIFT-Emulator serves to make emulation of SWIFT outputs very easy, while staying flexible enough to emulate anything, given a good set of training data.