Check whether HGF parameter estimation works using simulations where the ground truth is known
This is to replicate Figure 7 of Mathys et al. (2014) for the BFGS algorithm as implemented in the HGF Toolbox. The priors used are close to those in that paper, however some differences remain:
- omega is now omega_2
- kappa is now kappa_2 and is estimated in log-space instead of logit-space
- theta is now exp(omega_3), which is estimated in log-space
This is intended to run on ETH's Euler cluster. However, with a few straightforward modifications to simulate_estimate.sh, it can be run on any platform that has Bash and Matlab.
Entering
$ ./simulate_estimate.sh sims 256
at the shell prompt will give you 256 estimations per cell of your zeta/kappa grid, structured in a directory tree under the top-level directory sims. The first level of the tree contains directories named by zeta value, the second level by kappa value. The files containing the estimated kappa value are called ka{1..256}. A log file for each kappa's job is provided in job{1..256}.log. In order to have all estimated kappas in one text file for each grid cell, you can run
$ ./gather_kas.sh sims
This allows you to continue in Matlab by running (in the same directory)
>> gather_kas('sims');
>> load('kas.mat');
>> plot_kas(kas);
This gives you a plot of your estimates like this one here:
