Alternate to cell array

I am working with some codes and My traditional method has been to intialize an array and than put the output of the data into an array, I am doing this for matlabs classification learner and i made the mistake of using an array and hence the output is treated as induvidual features instead of one cell for each feature

clc; clear;

%% Load and truncate the signals
SigA_full = load("F2N_SR_S4_235.mat");
SigB_full = load("F2N_SR_S8_235.mat");

SigA = SigA_full.F2N_SR_S4_235(1:100000, :);  % Truncate SigA
SigB = SigB_full.F2N_SR_S8_235(1:100000, :);  % Truncate SigB

    %% High-pass Filter
    fs = 34000; % Hz
    filter_freq = 1000; % adjust as per your requirement
    filter_freq_norm = filter_freq/(fs/2);
    siglen = 50;    
    p = 1;
    for lidx = 1:siglen
            %% Load the signal1 and signal2
            signal1 = SigA(:, lidx)';
            signal2 = SigB(:, lidx)';          
            %% High-pass Filter
            signal1 = highpass(signal1, filter_freq_norm);
            signal2 = highpass(signal2, filter_freq_norm);
            signal = {signal1, signal2};
            signal = signal{1};
            %% Calculate features for each signal 
            % Kurtosis
            kurtosisValue = kurtosis(signal);
            % Mean
            meanValue = mean(signal);
            % Standard deviation
            sdValue = std(signal);
            % Variance
            varianceValue = var(signal);
            % Skewness
            skewnessValue = skewness(signal);
            %% Feature array Aggregation
            num_bins = 50; % Set bins 
            bin_averages_signal1 = process_signal(signal1, num_bins);
            bin_averages_signal2 = process_signal(signal2, num_bins);
     

    %% Create feature array
    features = [kurtosisValue, meanValue, sdValue, varianceValue, skewnessValue, bin_averages_signal1, bin_averages_signal2];

    %% Assign the calculated features to the array outside of the loop
    f2n3sw2(p,:) = features;
    p = p + 1;
    end

    %%
        function bin_averages = process_signal(signal, num_bins)
    % Find peaks and their heights
    [~, peak_heights] = findpeaks(signal);
    % Differences in height between adjacent peaks
    peak_height_diffs = diff(peak_heights);
    % Calculate the bin size
    bin_size = floor(length(peak_height_diffs) / num_bins);
    % Array to store the average values of each bin
    bin_averages = zeros(1, num_bins);
    % Loop through the peak height differences, calculate the average for each bin, and store it in the bin_averages array
    for i = 1:num_bins
        start_idx = (i - 1) * bin_size + 1;
        end_idx = i * bin_size;
        bin_data = peak_height_diffs(start_idx:end_idx);
        bin_averages(i) = mean(bin_data);
    end
   end

in this example as you can see me putting the features in a array the feature list is ok until bin_averages_signal1, bin_averages_signal2 where based on the bins it is 50 new cells the issue with this is classfication learner will take it as 50 unique features.

An example for what i want to do is something like this

It would be ok to run the code separately but i would prefer the feature files to be one file instead of having to combine them later as it would be ineffecicent.

[Image Analyst]

I'm having a little bit of trouble following. What variable or columns of variables are your predictors, and what vector is your response (ground truth classification) variable? Please show this screenshot after you have loaded up the variables from your workspace.

[Ammar]

@Image Analyst I am rusty in utilizing the classfication learner and machinelearning overall, so i use another code to do the labelling and label what the predictoes represent. As I have too many files, each with a different use case.

the code that has been sent is use for feature extraction and than in another version i combine them with my other data files and give them label such motion or very near motion. As for labelling the predictors i dont label and instead leave them as is, this is where i realized my mistake.

My feature array that i generated such as the example of the 1s and 0s is a custom code designed to get a binary output from 2 signals but it has lenght of 500, 1s and 0s, and as such when inputting into the classfication learner i realized they need to be put in one cell or else it is taken as 500 induvidual predictors which based of doing a weights analysis i relaized is not the right approach I had assumed that the predictors would be learning the pattern of the binary output.

same goes for the code provided above

features = [kurtosisValue, meanValue, sdValue, varianceValue, skewnessValue, bin_averages_signal1, bin_averages_signal2];

where while not lableded train feature 1 - 5 without labelling the predictors i knew what they were, and bin_averages_signal1 would represent values 6 to 55. But now i feel its the wrong way to do it and the predictor should be in one cell and not 6 to 55 hence i wanted to figure it out.

attached below is the way i combine the files and assign labels to them

BigFeatureArrayVNM = [vnm1sw1; vnm2sw1; vnm3sw1; vnm4sw1; vnm5sw1; vnm6sw1; vnm7sw1; vnm8sw1; vnm9sw1; vnm10sw1];

Unrecognized function or variable 'vnm1sw1'.

BigFeatureArrayMo = [n2f1sw1; n2f2sw1; n2f3sw1; n2f4sw1; n2f5sw1; fm1sw1; fm2sw1; f2n1sw1; f2n2sw1; f2n3sw1];
siglen = 500;
VNM = cellstr('HH');
for l = 1:siglen
    VNM(:,l) = cellstr('Very Near motion');
end
Mo = cellstr('HH');
for l = 1:siglen
    Mo(:,l) = cellstr('Motion');
end
FinalFeatVNM.Sig = BigFeatureArrayVNM;
FinalFeatVNM.Labels = VNM; 
FinalFeatMo.Sig = BigFeatureArrayMo;
FinalFeatMo.Labels = Mo;
TrainArr.Sig = [BigFeatureArrayVNM  BigFeatureArrayMo ];
TrainArr.Labels = [VNM Mo];
TrainFeatures = TrainArr.Sig';
trainingLabels = TrainArr.Labels';

AllFeats = [array2table(TrainFeatures) array2table(trainingLabels)];

Answer 1

[Shivansh]

Hi Ammar,

I understand that you are working on a classification problem and the output results are not correct because of labelling of predictors in dataset. It seems that you want to combine multiple feature arrays with their corresponding labels into a single training dataset for the classification learner.

You can use the following steps to write a script for creating your dataset:

1. Initialize empty arrays for ‘TrainFeatures’ and ‘trainingLabels’ before the loop.
2. Inside the loop, concatenate the current feature array with ‘TrainFeatures’ using the vertical concatenation operator ‘vertcat’.
3. Concatenate the current labels array with ‘trainingLabels’ using ‘vertcat’ as well.
4. After the loop, create a table from ‘TrainFeatures’ using ‘array2table’.
5. Concatenate ‘TrainFeaturesTable’ with ‘trainingLabels’ using ‘horzcat’ to create the final training dataset:

The AllFeats contains the combined feature arrays and their corresponding labels in a format suitable for training with the classification learner.

Here's the updated code snippet:

TrainFeatures = [];
trainingLabels = [];

% Loop through each feature array and labels
for i = 1:num_files
    currentFeatureArray = % Load the current feature array
    currentLabels = % Load the corresponding labels
    
    % Concatenate current feature array with TrainFeatures
    TrainFeatures = vertcat(TrainFeatures, currentFeatureArray);
    
    % Concatenate current labels with trainingLabels
    trainingLabels = vertcat(trainingLabels, currentLabels);
    
end

% Convert TrainFeatures to a table
TrainFeaturesTable = array2table(TrainFeatures);

% Concatenate TrainFeaturesTable with trainingLabels
AllFeats = horzcat(TrainFeaturesTable, array2table(trainingLabels));

You can use the ‘AllFeats’ table as your training dataset in the classification learner, where the features and labels are properly combined and organized.

You can refer to following documentations for the functions used in the code:

1. vertcat: Concatenate arrays vertically - MATLAB vertcat - MathWorks India
2. horzcat: Concatenate arrays horizontally - MATLAB horzcat - MathWorks India
3. array2table: Convert homogeneous array to table - MATLAB array2table - MathWorks India

Hope it helps!