error in tudadecode

[yingjoeyzhou]

I tried to train a STRM on some localizer trials, and apply the trained STRM on main trials of interest. However, i kept getting the error below while running tudadecode.m.

The error can be reproduced by evaluating the code below (the majority of code is copy-pasted from examples/STRM.m so I'd assume that the tudatrain.m is properly set up.

% ================== COPY-PASTE FROM `STRM.M` EXAMPLE STARTS =====================
%% PART ONE: Simulations for STRM-Classification model:
clear all;
rng(1);
P = 2; % number of data dimensions
Q = 2; % number of classes
K = 2; % number of latent states
T = 50; % number of timepoint in each trial
N = 20; % number of trials

% generate state timecourses:
z = zeros(N*T,K);
for itrial = 1:N
    transtime = rand(1)*48 + 1;
    t_offset = (itrial-1)*T;
    z(t_offset + [1:floor(transtime)],1)=1;
    z(t_offset + [ceil(transtime):T],2)=1;
end
cols = {[1 0 0],[0 1 0]}
% generate each class activation patterns and covariance matrices:
for ik=1:K
    W{ik} = randn(P+1,Q);
    Sigma{ik} = 0.5*wishrnd(eye(P),P);
end

% generate design matrices:
DM = [ones(N*T,1),[ones(N*T/2,1);zeros(N*T/2,1)],[zeros(N*T/2,1);ones(N*T/2,1)]];

% generate data itself:
data = zeros(N*T,P);
for itrial=1:N
    t_offset = (itrial-1)*T;
    for iT = 1:T
        data(t_offset+iT,:) = mvnrnd(DM(t_offset+iT,:)*W{find(z(t_offset+iT,:),1)},Sigma{find(z(t_offset+iT,:),1)});
    end
end

options = [];
options.K = K;
options.encodemodel = true; % this activates the STRM model
options.intercept = true;
options.covtype = 'full';
[STRMmodel,STRMstates] = tucatrain(data,DM(:,2:3),repmat(T,N,1),options);
% ================== COPY-PASTE FROM `STRM.M` EXAMPLE ENDS =====================


% decode
[Gamma,vpath] = tudadecode(data,DM,repmat(T,N,1),STRMmodel);

and the error message I get (in matlab2020b) looks like this:

I'm not sure if I had missed something in the pipeline, or it is something wrong with the sub-functions called by tudadecode?

Thanks in advance for your help!

[vidaurre]

Hi

In general, we stopped supporting the Matlab Toolbox, in favour of the Python one: https://github.com/vidaurre/glhmm
This can do all that the Matlab can do, and more: https://arxiv.org/pdf/2312.07151

(or OSL-dynamics, depending on the question and needs you have)

Thanks so much