Fully connected neural network invalid training data: Responses must be categorical sequences

Hello

I am trying to create a Fully Connected Neural Network to classify feature data. I have input data as a 4000x2 cell containing in column 1 the input features as 12x5 double (12 features with 5 values for each), and in column 2 the response lables (0 or 1). Before I input data to the network I arrange column 1 and 2 in seperate variables as a x = 4000x1 cell (with features) and y = 4000x1 categorical (with responses). However, when I run trainNetwork I get the following error message:

Error using trainNetwork (line 165)
Invalid training data. If the network outputs sequences, then the responses must be a cell array of categorical sequences, or
a categorical sequence.

I arrange and convert resonses from the input cell (4000x2) to a categorical by using categorical :

yTemp = inputDataCell(1:4000,2);
y = categorical(cell2mat(yTemp));

Question: How is my cell array of categorical responses not a categorical sequence and how do I make it so?

Answer 1

[Dams]

I found my own answer.

My fault was that I did in fact not give the network the responses as a sequence, as I thought. (surprise)

The network recieved input training data as a 4000x1 cell of 12x5 matrices and responses as a 4000x1 cell of 1x1 categorical lables (0 or 1). As the input has dimension 12 for each of the features and a sequence of 5, the response must also have a sequence of 5, hence the response 4000x1 cell should contain 1x5 categorical.

I fixed the response with a for loop adding the value of the response to a matrix from column 1 to 5:

yTemp = inputDataCell(1:4000,2);
yCategory = categorical(cell2mat(yTemp));
for i = 1:length(yCategory)
    tempClassNum = yCategory(i,1);
    tempClassMatrix(1,1:5) = tempClassNum;
    yCC(i) = {tempClassMatrix(:,:)};
end
ySequence = yCC';

[Andrew B]

I am still having an issue with my code. Can you check the thread that i posted?

[Javier Castillo]

I encountered the same problem, although this solution did not resolve the error as it would still show "Invalid training data. Responses must be a cell array of categorical response sequences." May I see your network structure? As my training data works on sequence to label networks, but not on my CNN.

[Vitor]

But this doesn't make any sense. If we consider this: https://www.mathworks.com/matlabcentral/answers/860840-could-anyone-help-me-how-to-solve-the-error-stating-y-must-be-a-vector-of-categorical-responses-or-a#comment_1594535 then the responses vector must be the same length as the number of cells of the input cell array.

As Walter said: "YTrain corresponds to class labels: each training sample (no matter how many features it has) corresponds to exactly one class."