AutoEmulate: A package for semi-automated emulation#
AutoEmulate is designed to be an easy, low-code pipeline to create emulators for complex simulations. At it’s core, AutoEmulate’s compare() function implements a typical machine learning workflow including data processing, selecting a good emulator model, optimising model parameters and evaluating its test-set performance. Emulators range from classical models like Radial Basis Functions and Second Order Polynomials to popular machine learning methods like Gradient Boosting and Support Vector Machines, as well as modern PyTorch-based models like Neural Processes and Multitask Gaussian Processes. All default parameters and search spaces for hyperparameter optimisation are chosen to be appropriate for typical emulation problems, i.e. small-ish datasets (100s or 1000s of datapoints) with potentially many features and outputs. We’ve also implemented Global Sensitivity Analysis as a common use-case for emulators, and plan to add more applications in the future.
Useful links: Code repository | Issues |