Need help with a for loop implementation for the below code

Greetings! I'm currently working on a project where in I perform a number of operations on a stimulus file (fancy name for a .mat file). The file consists of multiple trials and as of now, the following code is able to successfully carry out the process one trial at a time. I tried automating this code for all trials with the following but throws out this particular error:

N-dimensional indexing allowed for full matrices only

Can anyone help me debug the code. It's been bugging (!) me all day and I'm still not sure where I'm going wrong. Here's the code for reference:

function map = mkSaliencyMap(d,trialIds,varargin)
binsz = 0.1; 
sdev = 2.0;  
dt = 0.020;

    for itrial = (trialIds)
          
          if nargin < 3
              trialIds = 1:d.numTrials;
          end
          
          eyedata = d.meta.image.filename('time',Inf).data;
          eye = d.eye(trialIds(itrial));
          t0 = d.meta.cic.firstFrame('trial',trialIds(itrial));
          
          % Determining the start and end times at which the image is displayed
          
          [tstarttime,tr,frame,dta] = d.meta.image.startTime('trial',trialIds(itrial));
          [tstoptime, tr, frame, dta] = d.meta.image.stopTime('trial',trialIds(itrial));
          
          % Removing eyetraces before and after the image is displayed
          
          blink = false(size(eye.t));
          blink = blink | eye.t < tstarttime-t0;
          blink = blink | eye.t > tstoptime-t0;
          
          % Defining deformities in the plot as a result of blinks
          
          [tstart,tr,frame,dta] = d.meta.edf.STARTBLINK('trial',trialIds(itrial));
          tend = d.meta.edf.ENDBLINK('trial',trialIds(itrial));
          
          % Iterating over each blink datapoint
          
          tstart = tstart - t0 - dt
          tend = tend - t0 + dt
          
          if max(tend) < max(tstart)
              tend = [tend, Inf];
          end
          
          if min(tstart) > min(tend) % must be missing a start event
              tstart = [-Inf, tstart];
          end
          assert(numel(tstart) == numel(tend),'STARTBLINK/ENDBLINK event mismatch!');
          
          for itrial = 1:length(tstart)
              blink = blink | eye.t > tstart(itrial) & eye.t < tend(itrial);
          end
          
          % Defining the original dimensions of the image
          
          width = d.meta.image.width('trial', trialIds(itrial), 'time', Inf).data
          height = d.meta.image.height('trial', trialIds(itrial), 'time', Inf).data
          sz = ceil([height, width]./binsz);      % size of the saliency map
          x0 = floor(sz(2)/2)+1;
          y0 = floor(sz(1)/2)+1;                  % Co-ordinates of the screen centre
          
          % Removing eyetraces which are negative integers
          
          blink = blink | (ceil(eye.x/binsz)+x0 <= 0) | (ceil(eye.x/binsz)+x0 >= sz(2));
          blink = blink | (ceil(eye.y/binsz)+y0 <= 0) | (ceil(eye.y/binsz)+y0 >= sz(1));
          
          % Assigning NaN values to all blink datapoints for trial 1
          
          eye.x(blink) = NaN;
          eye.y(blink) = NaN;
          
          % Creating a sparse matrix of the eyetraces with the blinks removed
          
          map(:,:,itrial) = sparse(ceil(eye.y(~blink)/binsz)+y0,ceil(eye.x(~blink)/binsz)+x0,1,sz(1),sz(2));
          map(:,:,itrial) = full(map(:,:,itrial));
          map(:,:,itrial) = imgaussfilt(map(:,:,itrial),sdev);   % Gaussian filter with stdev = 2
      
      end
    end

[Walter Roberson]

Which line is generating the error message?

[Kevin Akash Rajasekaran]

The last part part of the code dealing with the sparse matrix maps i.e

map(:,:,itrial) = sparse(ceil(eye.y(~blink)/binsz)+y0,ceil(eye.x(~blink)/binsz)+x0,1,sz(1),sz(2));

[Walter Roberson]

What class() are the inputs?

At the time of the error, what are size() of each of the variables mentioned on the line ?

Answer 1

[Jan]

Multi-dim sparse arrays are not implemented in Matlab. Try to omit the sparse() .

Or use https://www.mathworks.com/matlabcentral/fileexchange/29832-n-dimensional-sparse-arrays

Answer 2

[Walter Roberson]

map(:,:,itrial) = sparse(ceil(eye.y(~blink)/binsz)+y0,ceil(eye.x(~blink)/binsz)+x0,1,sz(1),sz(2));

The right hand side appears to be a normal sparse array. However, sparse arrays cannot be assigned to a three dimensional array. For example

>> map(:,:,1) = sparse(1,1,1)
N-dimensional indexing allowed for full matrices only.

You will need to recode, such as

map(:,:,itrial) = full(sparse(ceil(eye.y(~blink)/binsz)+y0,ceil(eye.x(~blink)/binsz)+x0,1,sz(1),sz(2)));

in which case you can skip the next line

map(:,:,itrial) = full(map(:,:,itrial));   %no longer needed

You cannot mix sparse and full in the same array, by the way.