incorrect matrix size while training data

I am training a neural network with input [25x1x1].

I am taking input using imageInputLayer([25 1]).

My training data is sotred in the variable images which is of size [25x1x1x80000].

But when I run the program to train the network I get an error:

I dont know what is causing this error. My input dimensions match with the dimensions of my training data and everything else in the network seems fine.

Please help.

[KSSV]

Try squeezing the training data to 25*80000

[Arpan Parikh]

Tried that.

It gives error:

The training images are of size 80000x25x1 but the input layer expects images of size 25x1x1.

[Christian]

Mayge I get your question wrong, but this seems to be a problem of your order of dimensions. I never used it, but you could try to play around with permute().

E.g.:

image = images(1,:,:); % gives size [1x25x1]
image = permute(image, [2,1,3]); % or [2,3,1]

Make sure the image is not mirrored or rotated afterwards.

[Arpan Parikh]

Now I am getting the error:

The training images are of size 1x25x1 but the input layer expects images of size 25x1x1.

or

The training images are of size 1x1x25 but the input layer expects images of size 25x1x1.

[Arpan Parikh]

It seems that the dimensions are right but matlab is unable to extract training samples for some reason

Answer 1

[Raynier Suresh]

Hi, check whether you have defined the architecture of your network correctly, you can do this by using the command "analyzeNetwork(layers)". For input layer size [25 1] and the training data size [25 1 1 80000] the trainNetwork function should work fine for example you can refer the code below.

layers = [imageInputLayer([25 1])
    convolution2dLayer(1,32)
    reluLayer
    fullyConnectedLayer(4)
    softmaxLayer
    classificationLayer];

options = trainingOptions('sgdm', ...
    'MaxEpochs',1,...
    'InitialLearnRate',1e-4, ...
    'Verbose',false, ...
    'Plots','training-progress');

Xtrain = rand(25,1,1,80000);
Ytrain = categorical(randi(4,80000,1));

net = trainNetwork(Xtrain,Ytrain,layers,options);