Predictive inference for intractable models via ABC

This repository contains the code to reproduce the simulation experiments of the paper "On predictive inference for intractable models via approximate Bayesian computation". A preprint of the paper is available at: https://arxiv.org/abs/2203.12495.

The codebase also includes some additional features and simulation experiments (e.g. simplemarkov.inference). These are however not used in the paper and are not carefully tested.

To run the simulation experiments, the variable opt$root in simplemarkov1D.inference.R and in both *_setup.R-files need to be modified to point to the folder that contains the code files. Similarly, opt$save.loc needs to point to a folder where the computed results are to be saved.

It is also necessary to compile the C-codes of the simulation models using:

R CMD SHLIB simplemarkov_simul.c
R CMD SHLIB MG1_simul.c
R CMD SHLIB LV_simul.c

(The code has been tested using 64-bit Kubuntu 20.04 LTS with R version 3.6.3 "Holding the Windsock". On a Windows platform it might be necessary to additionally modify the dyn.load-command in each *_inference.R-file so that the resulting *.dll-files are loaded instead of the *.so-files used in Linux/OSX environments.)

The following R libraries are needed to run (some parts of) the code:

• matrixStats
• latex2exp
• coda

These libraries can be installed using install.packages().