How can I change a cell array with timepoints in to a continous binary matrix? (for Neural Network Toolbox)

I have a 15x1 cell array, in which each cell is another 50x2 cell. The data shows spike-event onsets in seconds for 50 neurons (each row is a neuron). The content looks like this:

    (XX{1,1}) = 
    {[  3.4078]}    {[  3.7273]}
    {0×0 double}    {0×0 double}
    {0×0 double}    {0×0 double}
    {[  3.4684]}    {0×0 double}
      ...

If this data is used for network training, this is the error that I get:

Error using trainNetwork (line 140)
Invalid training data. Predictors must be a cell array of sequences. The data dimension of all sequences must be the same.
Caused by:
  Error using nnet.internal.cnn.util.TrainNetworkDataValidator/assertValidSequenceInput (line 269)
  Invalid training data. Predictors must be a cell array of sequences. The data dimension of all sequences must be the same.

In order to use this as an input for a LSTM-Network in the Neural Network Toolbox these timepoints needs to be part of a continous spectrum. My idea was to convert this discrete datapoints in to large binary matrix (50x10000) inwhich the columns represent the time, therefore 10 seconds = 10000 columns. At each timepoint a spike occured (e.g. 3.4078) a 1 should be put for the corresponding neuron. I tried indexing the cell array with a time-vector ( t = (0:0.001:10) ) but it didn't work. Can anyone help transforming the data? Thank you.

Answer 1

[Mary Abbott]

Hello,

If I am understanding you correctly, you would like to change each 50x2 cell array into a 50x10000 matrix of mainly zeros, with a value of 1 at indices that correspond to the times in the original 50x2 array. You can do this by making a new 15x1 cell array in which each cell is a 50x10000 matrix of zeros. Then, use the times in the original array to index into the matrices of zeros to set the corresponding entry to 1. This would look something like:

% XX = original cell array
seqNum = 15;
neuronNum = 50;

% create new cell array
XX2 = cell([seqNum 1]);
for i = 1:seqNum
    XX2{i} = zeros(neuronNum, 10000);
    currSeq = XX{i};
    for j = 1:neuronNum
        % convert times into indices
        a = round(currSeq{j,1}*10000);
        b = round(currSeq{j,2}*10000);
        % Index into array of zeros
        XX2{i}(j, [a b]) = 1;
    end
end

[Burak Özbagci]

Thank you this did it!