Error using trainNetwork (line 191) TRANSPOSE does not support N-D arrays. Use PAGETRANSP​OSE/PAGECT​RANSPOSE to transpose pages or PERMUTE to reorder dimensions of N-D arrays

I'm trying to train a NN using 2000 sets of 3 x 128 data but getting error:

Error using trainNetwork (line 191)

TRANSPOSE does not support N-D arrays. Use PAGETRANSPOSE/PAGECTRANSPOSE to transpose pages or PERMUTE to reorder dimensions of N-D arrays.

Caused by:

Error using '

TRANSPOSE does not support N-D arrays. Use PAGETRANSPOSE/PAGECTRANSPOSE to transpose pages or PERMUTE to reorder dimensions of N-D arrays.

%here's my training data:
XTrain_arr=zeros(3,128,2000);
TTrain_arr=zeros(3,128,2000);
for i=1:2000
    XTrain_arr(:,:,i)=XTrain{i};
    TTrain_arr(:,:,i)=TTrain{i};
end
XTrain_arr=permute(XTrain_arr,[1,2,4,3]);
TTrain_arr=permute(TTrain_arr,[1,2,4,3]);
%defination of the network:
layers2 = [
    imageInputLayer([128, 1, 3], Name="input")
    convolution2dLayer([1, 4], 3, Padding="same", Stride=[1, 1])
    convolution2dLayer([1, 64], 8, Padding="same", Stride=[1, 8])    
    layerNormalizationLayer()
    scalingLayer(Scale=1, Offset=0)    
    reluLayer    
    maxPooling2dLayer([1, 2], Padding="same")
    convolution2dLayer([1, 32], 8, Padding="same", Stride=[1, 4])    
    layerNormalizationLayer()
    scalingLayer(Scale=1, Offset=0)
    reluLayer
    maxPooling2dLayer([1, 2], Padding="same")    
    transposedConv2dLayer([1, 32], 8, Cropping="same", Stride=[1, 4])
    reluLayer    
    transposedConv2dLayer([1, 64], 8, Cropping="same", Stride=[1, 8])
    reluLayer    
    flattenLayer()
    lstmLayer(8)
    fullyConnectedLayer(8)  
    dropoutLayer(0.2)
    fullyConnectedLayer(4)  
    dropoutLayer(0.2)
    fullyConnectedLayer(3)
    regressionLayer]
options = trainingOptions("adam",...
    MaxEpochs=600,...
    MiniBatchSize=32,...
    InitialLearnRate=0.001,...
    VerboseFrequency=100,...
    Verbose=1, ...
    Shuffle="every-epoch",...
    Plots="none", ...
    DispatchInBackground=true);
deepNetworkDesigner(layers2)
%Train the network
[net1_norm_2,info1_norm_2] = trainNetwork(XTrain_arr,TTrain_arr,layers2,options);

[Matt J]

%here's my training data:
XTrain_arr=zeros(3,128,2000);
TTrain_arr=zeros(3,128,2000);

XTrain_arr=permute(XTrain_arr,[1,2,4,3]);
TTrain_arr=permute(TTrain_arr,[1,2,4,3]);
%defination of the network:
layers2 = [
    imageInputLayer([128, 1, 3], Name="input")
    convolution2dLayer([1, 4], 3, Padding="same", Stride=[1, 1])
    convolution2dLayer([1, 64], 8, Padding="same", Stride=[1, 8])    
    layerNormalizationLayer()
    scalingLayer(Scale=1, Offset=0)    
    reluLayer    
    maxPooling2dLayer([1, 2], Padding="same")
    convolution2dLayer([1, 32], 8, Padding="same", Stride=[1, 4])    
    layerNormalizationLayer()
    scalingLayer(Scale=1, Offset=0)
    reluLayer
    maxPooling2dLayer([1, 2], Padding="same")    
    transposedConv2dLayer([1, 32], 8, Cropping="same", Stride=[1, 4])
    reluLayer    
    transposedConv2dLayer([1, 64], 8, Cropping="same", Stride=[1, 8])
    reluLayer    
    flattenLayer()
    lstmLayer(8)
    fullyConnectedLayer(8)  
    dropoutLayer(0.2)
    fullyConnectedLayer(4)  
    dropoutLayer(0.2)
    fullyConnectedLayer(3)
    regressionLayer]

layers2 = 
  24×1 Layer array with layers:

     1   'input'     Image Input                  128×1×3 images with 'zerocenter' normalization
     2   ''          2-D Convolution              3 1×4 convolutions with stride [1  1] and padding 'same'
     3   ''          2-D Convolution              8 1×64 convolutions with stride [1  8] and padding 'same'
     4   ''          Layer Normalization          Layer normalization
     5   'scaling'   Scaling                      Scaling
     6   ''          ReLU                         ReLU
     7   ''          2-D Max Pooling              1×2 max pooling with stride [1  1] and padding 'same'
     8   ''          2-D Convolution              8 1×32 convolutions with stride [1  4] and padding 'same'
     9   ''          Layer Normalization          Layer normalization
    10   'scaling'   Scaling                      Scaling
    11   ''          ReLU                         ReLU
    12   ''          2-D Max Pooling              1×2 max pooling with stride [1  1] and padding 'same'
    13   ''          2-D Transposed Convolution   8 1×32 transposed convolutions with stride [1  4] and cropping 'same'
    14   ''          ReLU                         ReLU
    15   ''          2-D Transposed Convolution   8 1×64 transposed convolutions with stride [1  8] and cropping 'same'
    16   ''          ReLU                         ReLU
    17   ''          Flatten                      Flatten
    18   ''          LSTM                         LSTM with 8 hidden units
    19   ''          Fully Connected              8 fully connected layer
    20   ''          Dropout                      20% dropout
    21   ''          Fully Connected              4 fully connected layer
    22   ''          Dropout                      20% dropout
    23   ''          Fully Connected              3 fully connected layer
    24   ''          Regression Output            mean-squared-error

options = trainingOptions("adam",...
    MaxEpochs=2,...
    MiniBatchSize=32,...
    InitialLearnRate=0.001,...
    Shuffle="every-epoch",...
    Plots="none");
%Train the network
[net1_norm_2,info1_norm_2] = trainNetwork(XTrain_arr,TTrain_arr,layers2,options)

Warning: Network issues detected.

Caused by:
Layer 10: Renamed. Layer was renamed 'scaling_2' because multiple layers had the name 'scaling'.
Layer 5: Renamed. Layer was renamed 'scaling_1' because multiple layers had the name 'scaling'.

Error using trainNetwork (line 191)
TRANSPOSE does not support N-D arrays. Use PAGETRANSPOSE/PAGECTRANSPOSE to transpose pages or PERMUTE to reorder dimensions of N-D arrays.

Caused by:
Error using '
TRANSPOSE does not support N-D arrays. Use PAGETRANSPOSE/PAGECTRANSPOSE to transpose pages or PERMUTE to reorder dimensions of N-D arrays.

Answer 1

[Hitesh]

Hi Ruoli,

The error indicates that the input data format 'XTrain_arr' or 'TTrain_arr' is incompatible with the expected format for "trainNetwork"."trainNetwork" expects input data 'XTrain_arr' to be formatted as a 4-D array in this format [height, width, channels, number of observations].

% Create dummy data for demonstration
XTrain = cell(1, 2000);
TTrain = cell(1, 2000);

for i = 1:2000
    XTrain{i} = rand(3, 128);       % 3 channels × 128 time steps
    TTrain{i} = rand(1, 3);         % Regression target: 1×3 vector
end

XTrain_arr = zeros(128, 1, 3, 2000);  % image format for imageInputLayer
TTrain_arr = zeros(2000, 3);         % regression targets

for i = 1:2000
    X = XTrain{i}';                  % Now X is 128×3
    XTrain_arr(:,1,:,i) = X;         % Format: H × W × C × N
    TTrain_arr(i,:) = TTrain{i};     % Format: N × output_dim
end

% Define the network (assuming you fixed the scalingLayer as discussed earlier)
layers2 = [
    imageInputLayer([128, 1, 3], Name="input")
    convolution2dLayer([1, 4], 3, Padding="same", Stride=[1, 1])
    convolution2dLayer([1, 64], 8, Padding="same", Stride=[1, 8])    
    layerNormalizationLayer()
    scalingLayer(Scale=1)    
    reluLayer    
    maxPooling2dLayer([1, 2], Padding="same")
    convolution2dLayer([1, 32], 8, Padding="same", Stride=[1, 4])    
    layerNormalizationLayer()
    scalingLayer(Scale=1)
    reluLayer
    maxPooling2dLayer([1, 2], Padding="same")    
    transposedConv2dLayer([1, 32], 8, Cropping="same", Stride=[1, 4])
    reluLayer    
    transposedConv2dLayer([1, 64], 8, Cropping="same", Stride=[1, 8])
    reluLayer    
    flattenLayer()
    lstmLayer(8)
    fullyConnectedLayer(8)  
    dropoutLayer(0.2)
    fullyConnectedLayer(4)  
    dropoutLayer(0.2)
    fullyConnectedLayer(3)
    regressionLayer
];

% Training options
options = trainingOptions("adam",...
    MaxEpochs=600,...
    MiniBatchSize=32,...
    InitialLearnRate=0.001,...
    VerboseFrequency=100,...
    Verbose=1, ...
    Shuffle="every-epoch",...
    Plots="none", ...
    DispatchInBackground=true);

% Train the network
[net1_norm_2, info1_norm_2] = trainNetwork(XTrain_arr, TTrain_arr, layers2, options);

However, "trainNetwork" is not recommended. Use the trainnet function instead as mentioned in MATALB documentation.

Answer 2

[Ruoli]

Hi Hitesh,

Thank you for your comment! I have a follow-up question regarding my training data. My input and output consist of 2000 samples of 3-channel 128 length signals. I'm structuring the training data like this:

XTrain_arr = zeros(128, 1, 3, 2000);
TTrain_arr = zeros(3, 128, 2000);
for i = 1:2000
    X = XTrain{i}';
    XTrain_arr(:,1,:,i) = X;
    TTrain_arr(:,:,i) = TTrain{i};
end

Then, I defined the network as you suggested and ran the trainNetwork command:

[net1_norm_2, info1_norm_2] = trainNetwork(XTrain_arr, TTrain_arr, layers2, options);

However, I still encountered the error:

Error using trainNetwork (line 191) TRANSPOSE does not support N-D arrays. Use PAGETRANSPOSE/PAGECTRANSPOSE to transpose pages or PERMUTE to reorder dimensions of N-D arrays. Caused by: Error using ' TRANSPOSE does not support N-D arrays. Use PAGETRANSPOSE/PAGECTRANSPOSE to transpose pages or PERMUTE to reorder dimensions of N-D arrays.

Could you please advise on how to resolve this error? Is there a specific way to format the data or adjust the network layers to avoid this issue?

Thank you!

[Hitesh]

Hi Ruoli,

Kindly update "TTrain_arr as below due to regression targets format and let me know if you still face this issue.

TTrain_arr = zeros(2000, 3);