Contact Us

Thanks for visiting the website. If you want to get in touch you can reach me through Twitter or please fill out the form on the right.

         

123 Street Avenue, City Town, 99999

(123) 555-6789

email@address.com

 

You can set your address, phone number, email and site description in the settings tab.
Link to read me page with more information.

"Model Dispersion with prism: An Alternative to MCMC for Rapid Analysis of Models" - vd Velden (2019)

Blog

"Model Dispersion with prism: An Alternative to MCMC for Rapid Analysis of Models" - vd Velden (2019)

Alan Duffy

Published in the highly ranked ApJS and JOSS (as well as being available on arXiv for free) the “Probabilistic Regression Instrument for Simulating Models” package PRISM was a massive undertaking by my PhD student Ellert vd Velden. He built a brand-new open source MPI-capable Python package that can take ANY model from a user and map out the entire parameter space for regions that can explain the data. PRISM is available on pypi (i.e. you can just “pip install prism”) and within minutes you can start exploring your model thanks to the insanely well documented guide.

PRISM uses an emulator technique to make a model of your model (i.e. explaining the full parameter space using a handful of polynomials) and automatically increasing the sampling where needed for regions that are yet to be faithfully predicted. It’s a beautiful and robust technique, with a codebase that is portable, MPI-capable and can be run alone or in a hybrid sampling mode with MCMC (essentially telling it where to evaluate chains) means it should be the default first call for any serious analysis effort.

A figure from the paper showing the parameter space (i.e. where the model can describe the data) before and after a PRISM evaluation. The dramatic shrinking of the parameter space allows a user to both completely understand their model (degeneracies and the like) as well as focus precious computational resources on exploring only the likely space on the right using more expensive techniques like MCMC.

A figure from the paper showing the parameter space (i.e. where the model can describe the data) before and after a PRISM evaluation. The dramatic shrinking of the parameter space allows a user to both completely understand their model (degeneracies and the like) as well as focus precious computational resources on exploring only the likely space on the right using more expensive techniques like MCMC.