Model dispersion with PRISM; an alternative to MCMC for rapid analysis of models. (arXiv:1901.08725v1 [astro-ph.IM])

Model dispersion with PRISM; an alternative to MCMC for rapid analysis of models. (arXiv:1901.08725v1 [astro-ph.IM])
<a href="http://arxiv.org/find/astro-ph/1/au:+Velden_E/0/1/0/all/0/1">Ellert van der Velden</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Duffy_A/0/1/0/all/0/1">Alan R. Duffy</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Croton_D/0/1/0/all/0/1">Darren Croton</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Mutch_S/0/1/0/all/0/1">Simon J. Mutch</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Sinha_M/0/1/0/all/0/1">Manodeep Sinha</a>

We have built PRISM, a “Probabilistic Regression Instrument for Simulating
Models”. PRISM uses the Bayes linear approach and history matching to construct
an approximation (’emulator’) of any given model, by combining limited model
evaluations with advanced regression techniques, covariances and probability
calculations. It is designed to easily facilitate and enhance existing Markov
chain Monte Carlo (MCMC) methods by restricting plausible regions and exploring
parameter space efficiently. However, PRISM can additionally be used as a
standalone alternative to MCMC for model analysis, providing insight into the
behavior of complex scientific models. With PRISM, the time spent on evaluating
a model is minimized, providing developers with an advanced model analysis for
a fraction of the time required by more traditional methods.

This paper provides an overview of the different techniques and algorithms
that are used within PRISM. We demonstrate the advantage of using the Bayes
linear approach over a full Bayesian analysis when analyzing complex models.
Our results show how much information can be captured by PRISM and how one can
combine it with MCMC methods to significantly speed up calibration processes
(>15 times faster). PRISM is an open-source Python package that is available
under the BSD 3-Clause License (BSD-3) at https://github.com/1313e/PRISM and
hosted at https://prism-tool.readthedocs.io

We have built PRISM, a “Probabilistic Regression Instrument for Simulating
Models”. PRISM uses the Bayes linear approach and history matching to construct
an approximation (’emulator’) of any given model, by combining limited model
evaluations with advanced regression techniques, covariances and probability
calculations. It is designed to easily facilitate and enhance existing Markov
chain Monte Carlo (MCMC) methods by restricting plausible regions and exploring
parameter space efficiently. However, PRISM can additionally be used as a
standalone alternative to MCMC for model analysis, providing insight into the
behavior of complex scientific models. With PRISM, the time spent on evaluating
a model is minimized, providing developers with an advanced model analysis for
a fraction of the time required by more traditional methods.

This paper provides an overview of the different techniques and algorithms
that are used within PRISM. We demonstrate the advantage of using the Bayes
linear approach over a full Bayesian analysis when analyzing complex models.
Our results show how much information can be captured by PRISM and how one can
combine it with MCMC methods to significantly speed up calibration processes
(>15 times faster). PRISM is an open-source Python package that is available
under the BSD 3-Clause License (BSD-3) at https://github.com/1313e/PRISM and
hosted at https://prism-tool.readthedocs.io

http://arxiv.org/icons/sfx.gif