Introduction#
This collection of notebooks aims to provide the basic yet practical usage of surmise in constructing statistical emulators and conducting Bayesian calibration. In a nutshell, Bayesian calibration refers to the task of learning unknown parameters of a simulation model through estimating the posterior distribution of the parameters. When a simulation model is computationally expensive, emulators serve as fast approximations that are constructed based on simulation outputs.
Readers may use these notebooks for several purposes. First and foremost, the notebooks are examples for learning the functionalities offered in surmise. Second, the notebooks are case studies of general topics in Bayesian calibration, for which each notebook introduces the context and topics it covers.
For an in-depth introduction of Bayesian calibration, readers may refer to Kennedy and O'Hagan [2001]. For exploring the basics of Gaussian process emulators, which form the foundation of many available emulators in surmise, readers are directed to Gramacy [2020], Santner et al. [2003], Williams and Rasmussen [2006].
Documentation#
The full documentation including the programmatic interface of the package is available in surmise’s user and developer guides.
Interactivity#
The notebooks included in this book can be launched via Binder so that users can run them interactively. Please note that the notebooks often contain cells that are not present in the book’s rendering. These cells can contain
commented out lines that specify other emulator, calibrator, or sampler methods that could be used in the notebook or
code that checks correct execution of the notebook.
The latter can be ignored since they are for development and maintenance purposes only.
Table of contents#
Examples
Reference