Boosting MonteCarlo sampling with a non-Gaussian fit
arXiv:2007.02615 · doi:10.1093/mnras/staa2362
Abstract
We propose a new method, called MonteCarlo Posterior Fit, to boost the MonteCarlo sampling of likelihood (posterior) functions. The idea is to approximate the posterior function by an analytical multidimensional non-Gaussian fit. The many free parameters of this fit can be obtained by a smaller sampling than is needed to derive the full numerical posterior. In the examples that we consider, based on supernovae and cosmic microwave background data, we find that one needs an order of magnitude smaller sampling than in the standard algorithms to achieve comparable precision. This method can be applied to a variety of situations and is expected to significantly improve the performance of the MonteCarlo routines in all the cases in which sampling is very time-consuming. Finally, it can also be applied to Fisher matrix forecasts, and can help solve various limitations of the standard approach.
arXiv comment14 pages, 8 figures; References, comments, and appendix B added; The MCPostFit code can be found in: https://github.com/adriagova/MCPostFit
References in corpus (13)
- The Complete Light-curve Sample of Spectroscopically Confirmed Type Ia Supernovae from Pan-STARRS1 and Cosmological Constraints from The Combined Pantheon Sample
- The Cosmic Linear Anisotropy Solving System (CLASS) II: Approximation schemes
- Cosmology and Fundamental Physics with the Euclid Satellite
- Conservative Constraints on Early Cosmology: an illustration of the Monte Python cosmological parameter inference code
- CDM Models with a Smooth Component
- Type Ia Supernova Distances at z > 1.5 from the Hubble Space Telescope Multi-Cycle Treasury Programs: The Early Expansion Rate
- Approximate Bayesian Computation for Forward Modeling in Cosmology
- cosmoabc: Likelihood-free inference via Population Monte Carlo Approximate Bayesian Computation
- Bayesian optimisation for likelihood-free cosmological inference
- Breaking the spell of Gaussianity: forecasting with higher order Fisher matrices
- Cosmological parameter estimation via iterative emulation of likelihoods
- A fast, always positive definite and normalizable approximation of non-Gaussian likelihoods
- Reconstructing Probability Distributions with Gaussian Processes