stratABC

ABC using stratified Monte Carlo and bootstrapping: supporting code for the paper by U. Picchini and R.G. Everitt "Stratified sampling and bootstrapping for approximate Bayesian computation", arXiv:1905.07976

The following folders are included:

• "gauss", illustrating the first case study in arXiv:1905.07976, MATLAB code.
• "gk", illustrating the a case study in Supplementary Material of arXiv:1905.07976, MATLAB and R code.
• "supernova", illustrating a case study in arXiv:1905.07976, MATLAB code.
• "lotka-volterra", illustrating a case study in arXiv:1905.07976, MATLAB code.

Here we describe the content of each folder:

• "gauss"
  • demo_gauss.m: runs pmABC-MCMC
  • demo_gauss_resampling.m: runs rABC-MCMC
  • demo_gauss_stratified_3strata.m: runs rsABC-MCMC
  • demo_gauss_stratified_3strata_averagedlikelihoods.m: runs xrsABC-MCMC
  • demo_gauss_two_independent_sample: runs pmABC-MCMC with M=2
  • subfolder "loglikelihood estimation": produces results for section 6.1.1
  • subfolder "appendix_code": produces results for the Supplementary Material section "Efficiency of the averaged likelihood approach"
• "gk"
  • subfolder "stratifiedABC" performs a few iterations with rABC-MCMC and then a few more using rsABC-MCMC follow.
  • subfolder "exchanged-likelihoods" performs a few iterations with rABC-MCMC and then a few more using xrsABC-MCMC follow.
  • subfolder "ABCmultiplesamples" performs pmABC-MCMC.
• "lotka-volterra"
  • subfolder "ABC-SMC" runs sequential Monte Carlo ABC (no resamplig, no stratification) using the algorithm described in Supplementary Material
  • subfolder "pseudomarginalABC_threshold=0.6" runs pmABC-MCMC
  • subfolder "rsABC-MCMC_3strata" runs rsABC-MCMC using three strata.
  • subfolder "several-bootstrap-comparisons" contains results as given in Supplementary Material, comparing the performance of several bootstrap strategies.
  • subfolder "computationally-intensive-model": contains runs of pmABCMCMC and rsABCMCMC for the expensive case study considered in Supplementary Material.
• "supernova"
  • subfolder "importancesampling" runs importance sampling ABC with and without stratification. Different RUN files are provided ("astro_run_stratified" uses stratified Monte Carlo and astro_run_nostratification_M_1.m and astro_run_nostratification_M_2.m do not use stratified MC).
  • subfolder "rejectionABC" uses the simple ABC-rejection algorithm, with and without stratification. Different RUN files are provided, same as for the "importancesampling" subfolder