CNN: How to create label vector acceptable by trainNetwork

I'm trying to create cnn for load forecasting. I try to create lebel vector for training but it didn't work.

I have 3 types of data with 1920 samples fo each for training set (1920x3 double) and 3 types of data with 768 samples fo each for testing set (768x3 double). I reshaped train data to 4D array.

this is my code for this work.

%% Reshaped input
Train_dataset = reshape(Train_data, [1 1920 1 3]);

%  Creat the labels
label = (1:3,:);
Train_Labels = categorical(label); 


%% Layers
layers = [ ...
    imageInputLayer([1 1920 1])
    convolution2dLayer(1,20)
    reluLayer
    maxPooling2dLayer(1,'Stride',1)
    fullyConnectedLayer(1)
    softmaxLayer
    classificationLayer];


%% Train Options 
train_options = trainingOptions('sgdm');


%% Train network
net = trainNetwork(Train_dataset, Train_Labels, layers, train_options);


%% Test network

Pred = classify(net, Test_data)'

What is the train lebel should look like? or somewhere was wrong?

the error frequently shows 'Number of observations in X and Y disagree.'

Answer 1

[Anshika Chaurasia]

Hi,

The error is appearing because the number of observations in "Train_dataset" and "Train_Labels" are different.

So, if "Train_dataset" is 1920x3 then "Train_Labels" should be 1920x1.

One more thing, following line can throw error as indexing can be applied on an array:

label = (1:3,:);