How to run this CNN?

Hello. Good time. According to the Matlab code I wrote, I want to consider the position, velocity, and acceleration features together and feed them to the convolutional network. I get an error. The first column of my data is time and the rest of the columns are features. The data is related to the skeletal position of the 25 joints of the body. I uploaded the workspace and Matlab code. Please help. Error: Invalid training data. Predictors must be a numeric array, a datastore, or a table. For networks with sequence input, predictors can also be a cell array of sequences.

[Pratyush Swain]

Hi NGR,

I am unable to extract the workspace folder zip, it shows error stating the zip file is empty. Can you upload it again or upload any other relevant data files required to run your script ?

Thanks

Answer 1

[Pratyush Swain]

Hi NGR,

In your provided script, the dimensions were managed by resizing entire data arrays to match a target row size 'desiredRows'. This approach maintained the data structure class wise which might have led to complications while splitting data for training/testing and reshaping for input into the network.

Alternatively you can try this approach where the data is padded to have a fixed set of rows for each class - 'maxRows' and each row of the padded data is treated as a separate sample, so the final structure of 'XTrainArray' has dimensions [maxRows, desiredCols, channels, total_samples] where:

• 'maxRows' is the padded row size.
• 'desiredCols' is 72, the fixed number of features (columns).
• 'channels' is 3 (position, velocity, acceleration).
• 'total_samples' is the number of samples from all classes.

Please refer to this example implementation:

% Calculate maxRows (max number of rows in the data under each class)
maxRows = max(cellfun(@(x) size(x, 1), positionData_resized));

% Initialize XTrain and YTrain
XTrain = {};
YTrain = [];

% Prepare data and labels
for i = 1:numCells
    % Concatenate position, velocity, acceleration
    data = cat(3, positionData_resized{i}, velocityData_resized{i}, accelerationData_resized{i});
    
    % Pad data with zeros to match maxRows
    paddedData = padarray(data, [maxRows - size(data, 1), 0, 0], 0, 'post');
    
    % Append each row (slice) of paddedData as a separate sample to XTrain
    for j = 1:maxRows
        XTrain{end+1} = paddedData(j, :, :); % Each cell contains [1, 72, 3]
        YTrain = [YTrain; i]; % Label for each sample
    end
end

% Convert XTrain to a 4D array
XTrainArray = cat(4, XTrain{:});

% Convert YTrain to categorical
YTrain = categorical(YTrain);

% Adjusted input size
inputSize = [1, 72, 3];
layers = [
    imageInputLayer(inputSize, 'Normalization', 'none', 'Name', 'Input')
    convolution2dLayer([3 3], 32, 'Padding', 'same', 'Name', 'Conv1')
    batchNormalizationLayer('Name', 'BatchNorm1')
    reluLayer('Name', 'ReLU1')
    fullyConnectedLayer(numCells, 'Name', 'FC')
    softmaxLayer('Name', 'Softmax')
    classificationLayer('Name', 'Output')];


% Training options
options = trainingOptions('sgdm', ...
    'MaxEpochs', 50, ...
    'InitialLearnRate', 0.01, ...
    'Verbose', false, ...
    'Plots', 'training-progress');

% Train the network
net = trainNetwork(XTrainArray, YTrain, layers, options);

You can prepare the test data similar to the training data in the above implementation.

For more information on 'trainNetwork' function, please refer https://www.mathworks.com/help/deeplearning/ref/trainnetwork.html

Hope this helps.