This repository contains a collection of Matlab files that contain code to accompany the paper "Factorial comparison of working memory models" by Van den Berg, Awh, and Ma (Psychological Review, 2014).
The code is quite extensively annotated. However, if you have any questions/bug reports/etc, please contact me at nronaldvdberg@gmail.com.
To get started, I'd recommend to run and look at run_demo.m.
--- FILE DESCRIPTIONS ---
run_demo.m : this walks you through generating synthetic data and fitting models to those data
fit_model.m : used to fit the 32 main models described in the the paper gen_fake_data: used to generate synthetic data from the 32 main models described in the paper (note that these functions were (re)written for clarity, not speed; considerable speeding up can be achieved by vectorizing the code)
get_gvar.m : return general settings for model fitting (mostly settings of the evolutionary optimization method) reproduce.m : draw parameter values from a specific distribution (used by the optimization algorithm)
besseli0_fast.m: evaluation I0(.) in a way faster than Matlab's besseli()
circ_*.m : these files are part of the Circular Statistics Toolbox by P. Berens and J. Velasco randi.m : should do the same as Matlab's randi.m (some older version of Matlab don't appear to have this function)
--- DATA STRUCTURE ---
The fake data that are returned by gen_fake_data and the input data for fit_model are structures with three fields: error_vec, dist_error_vec, and N.
N gives the set size on each trial.
The error_vec field contains the errors on all trials, i.e., the (circular) difference between the response and the target value. The domain is [-pi, pi], because that is the domain of the Von Mises distribution. If your errors are for example in the range [-180, 180], then you should rescale them to [-pi, pi], by multiplying them by pi/180.
The dist_error_vec contains the "non-target errors" on each trial, i.e., the (circular) difference between the response and each non-target value (again in the range [-pi, pi]). Say, for example, that we have N=4 on the first trial. Then dist_error_vec{1} contains 3 values, corresponding to the distances between the response and the first, second, and third non-target item, respectively. This information is only required when you fit models that include non-target responses, i.e., all the XXX-NT models. Introduction: https://www.cns.nyu.edu/malab/resources.html#DE
Also refer to readme in the code folder.