Abandoned

This project is abandoned - I would not recommend using it. I'd much rather recommend helping in developung Unfold.jl which has already support for MixedModels, albeit with some problems on realistic models (2021-11-09)

Disclaimer:

This extension is still being developed and needs optimization for speed and additional tests. Use at your own risk!. Probably best to contact me prior to using it to get to know the quirks :-)

Unfold + Mixed Models = unmixed

Deconvolution, non linear modeling and Mixed Modeling toolbox

Specify y~1 + A + (1 + A|subject) + (1+C|image) Item and Subject Effects!

Install

git clone https://github.com/behinger/unmixed
git submodule update --init --recursive

Minimal Example:

rng(1)
input = [];
for k = 1:30
    % you can also add item effects by the adding the flag 'randomItem', 1
    % in simulate_data_lmm. The betas / thetas are currently hardcoded because I'm lazy
    input{k} = simulate_data_lmm('noise',10,...
        'srate',50,'datasamples',600*50,...
        'basis','dirac');
end


EEG = um_designmat(input,'eventtypes','fixation','formula','y~1+A+(1+A|subject)');

EEG= um_timeexpandDesignmat(EEG,'timelimits',[-0.1,0.5]);

% Currently I recommend the bobyqa optimizer. Seems to be faster
model_fv = um_mmfit(EEG,input,'channel',1,'optimizer','bobyqa','covariance','FullCholeksy'); % Todo: Directly read covariance from formula like lme4