The sample-matlab-code folder contains two MATLAB implementations of the SAEM-SL methodology by Umberto Picchini, described in "Likelihood-free stochastic approximation EM for inference in complex models", arXiv:1609.03508.
Please refer also to the README files in each of the subfolders "nonlingauss" and "theophylline" for further information.
The software contained in the two subfolders calls the following utilities:
- neglogmvnpdf utility to compute the negative log pdf of a multivariate Gaussian
- nearestSPD utility to compute N. Higham's "nearest semi-positive definite matrix"
- param_mask utility: extract the free parameters from the full vector of free and fixed parameters
- param_unmask utility: reconstruct the full vector of structural model parameters from the set of free parameters
- hline utility that adds an horizontal line to plots