Abstract

Setting up a new simulator, or a new implementation of an existing one, requires lower and upper bounds for the ranges of each of the parameters that are not perfectly known. Getting these ranges wrong can lead to a very misleading sensitivity analysis, and many wasted simulator runs. Dr. Jony Rougier explain how 'implausibility' can be used to rule out poor choices of the parameter values using observations, and how this allows us to identify a rectangular region of 'not ruled out yet' (NROY) values, through an iterative process which makes efficient use of your time and your CPU cycles. If there is time he will also discuss why and how to transform fields of parameters (e.g. a 1D field of Manning's n values along a river channel, or a 3D field of porosity values in a hydrocarbon reservoir simulator).

The presentation is based around the paper ‘Setting Up Your Simulator’ (in submission to the Journal of Uncertainty Quantification), available here.

It is also accompanied by a worksheet in R, available here.

This worksheet contains all of the functions necessary to Set Up Your Simulator, as well as a demonstration of those functions in action, and visualisation of the results.

Tutorial structure

1.5 hour lead-discussion on the whys and hows of setting ranges for the parameters of a computer simulator, followed by a 1.5 hour hands-on session (bring your own laptop), following the SUYS worksheet (see above). 

Lunch will be provided. 

Background required

Some experience of modelling and computer simulation. The hands-on session requires a laptop with R installed. See here for details.

Registration

Numbers are limited so booking via Eventbrite is essential. 

References

Some of the issues are discussed in:

I. Vernon, M. Goldstein, and R.G. Bower, 2010. Galaxy formation: A Bayesian uncertainty analysis. Bayesian Analysis 5(4), 619-670. Available online (including discussion and rejoinder).