Error Invalid Training Data- Predictors must be a N-by-1 cell array of sequences?

I am unable to solve the error Invalid training data. Predictors must be a N-by-1 cell array of sequences, where N is the number of sequences. All sequences must have the same

% Assume 'cnnDataArray' contains CNN-extracted features and 'allLabels' has the corresponding labels.
% Validate that the size of the data matches the labels
%cnnDataArray = cnnDataArray(1:numel(allLabels), :);
% Validate and align data and labels
numSamplesData = size(cnnDataArray, 1);
numSamplesLabels = numel(allLabels);
if numSamplesData > numSamplesLabels
    fprintf('Truncating data to match labels.\n');
    cnnDataArray = cnnDataArray(1:numSamplesLabels, :);
elseif numSamplesLabels > numSamplesData
    fprintf('Truncating labels to match data.\n');
    allLabels = allLabels(1:numSamplesData);
end
% Check consistency
if size(cnnDataArray, 1) ~= numel(allLabels)
    error('Mismatch persists after alignment. Data samples: %d, Labels: %d.', size(cnnDataArray, 1), numel(allLabels));
end
disp('Data and labels are aligned.');
% Convert labels to categorical if not already
allLabels = allLabels(1:size(cnnDataArray, 1));
% Validate that lstmInput has the correct dimensions
[numSamples, ~, numFeatures] = size(lstmInput);
% Expanding trainDataCell
expandedTrainDataCell = cell(numel(trainDataCell), 1);  % Create a new cell array to hold the expanded sequences
for i = 1:numel(trainDataCell)
    % Ensure each sequence is correctly formatted
    if size(trainDataCell{i}, 1) == 1
        % If it's a single time step with F features, no need to reshape, keep it as [1, F]
        expandedTrainDataCell{i} = trainDataCell{i};  % Keep as is
    else
        % If there are multiple time steps, keep the sequence structure intact
        expandedTrainDataCell{i} = trainDataCell{i};  % Sequence remains as is
    end
end
% Expanding testDataCell similarly
expandedTestDataCell = cell(numel(testDataCell), 1);
for i = 1:numel(testDataCell)
    if size(testDataCell{i}, 1) == 1
        % If it's a single time step with F features, keep it as is
        expandedTestDataCell{i} = testDataCell{i};
    else
        % Otherwise, keep the sequence structure intact
        expandedTestDataCell{i} = testDataCell{i};
    end
end
% Check the size of the expanded cells
disp(size(expandedTrainDataCell));  % Should show [N, 1]
disp(size(expandedTestDataCell));   % Should show [M, 1]
% Now you can use these cell arrays directly for LSTM training
% Do not use cell2mat unless you need a matrix of fixed-size sequences
% Continue with training the LSTM using the expanded cell arrays
% Reshape data for LSTM
%lstmInput = reshape(cnnDataArray, [numSamples, 1, numFeatures]); % [numSamples, 1, numFeatures]
% Verify the reshaped data
%disp(size(lstmInput)); % Should display [numSamples, 1, numFeatures]
% Split data into training and testing sets (e.g., 80-20 split)
%cv = cvpartition(allLabels, 'Holdout', 0.2);  % Adjust the holdout ratio if necessary
%trainIdx = training(cv);

[Ayush Aniket]

Share the complete code you are working with including the layers of your neural network.

[Cris LaPierre]

And share the full error message (all the red text).

[PRACHI BHAGAT]

% Define CNN layers
layers = [
    % Input layer
    imageInputLayer([1 numFeatures 1], 'Name', 'input', 'Normalization', 'none')
    
    % Convolutional Layer 1
    convolution2dLayer([1 5], 32, 'Stride', 1, 'Padding', 'same', 'Name', 'conv1') % 32 filters
    batchNormalizationLayer('Name', 'batchnorm1')
    reluLayer('Name', 'relu1')
    
    % Convolutional Layer 2
    convolution2dLayer([1 5], 64, 'Stride', 1, 'Padding', 'same', 'Name', 'conv2') % 64 filters
    batchNormalizationLayer('Name', 'batchnorm2')
    reluLayer('Name', 'relu2')
    dropoutLayer(0.2, 'Name', 'dropout1') % Regularization with 20% dropout
    
    % Convolutional Layer 3
    convolution2dLayer([1 5], 128, 'Stride', 1, 'Padding', 'same', 'Name', 'conv3') % 128 filters
    batchNormalizationLayer('Name', 'batchnorm3')
    reluLayer('Name', 'relu3')
    dropoutLayer(0.3, 'Name', 'dropout2') % Regularization with 30% dropout
    
    % Fully Connected Layer 1
    fullyConnectedLayer(64, 'Name', 'fc1') % Dense layer with 64 neurons
    reluLayer('Name', 'relu_fc1')
    dropoutLayer(0.4, 'Name', 'dropout_fc1') % Regularization with 40% dropout
    
    % Fully Connected Layer 2
    fullyConnectedLayer(numel(unique(allLabels)), 'Name', 'fc2') % Output layer with neurons equal to the number of classes
    softmaxLayer('Name', 'softmax') % Softmax for classification
    classificationLayer('Name', 'output') % Classification output
    ];

This is the CNN layer

[PRACHI BHAGAT]

error using trainnetwork (line 191) invalid training data. predictors must be a n-by-1 cell array of sequences, where n is the number of sequences. all sequences must have the same feature dimension and at least one time step. error in cnn2_utdataset (line 207) net = trainnetwork(traindatacell, trainlabels, layers, options);

this is the error

Answer 1

[Ayush Aniket]

Based on the error message, the error occurs due to discrepancy between the expectda data format and the format of your data.

The trainNetwork function expects data for 2-D image sequences (as it seems from the information provided) to be in the format of Nx1 cell array where each element is a h-by-w-by-c-by-s arrays, where h, w, and c correspond to the height, width, and number of channels of the images, respectively, and s is the sequence length.

Hence, you training data traindatacell (predictors) must be in the following format: Nx1 cell array where each element is a [1 numFeatures 1 s] array.

Refer the documentation link below to read about the expected format for different type of data: https://www.mathworks.com/help/deeplearning/ref/trainnetwork.html?#mw_36a68d96-8505-4b8d-b338-44e1efa9cc5e

Note: From the code provided, the imageInputLayer expects input in the format [1 numFeatures 1]. However as mentioned in its documentation, the expected format is a row vector of integers [h w c], where h, w, and c correspond to the height, width, and number of channels respectively. Assuming, numfeatures to be the number of channels, you should modify the format to the layer to [1 1 numFeatures]. Refer the documentation here: https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.imageinputlayer.html#mw_342fa7c6-d7c0-456b-bfa5-366256fe67c9

If you are using any other data type, please share the format of the data that you are working with.

[PRACHI BHAGAT]

ok thanks Ayush