The indices for validation and test sets are not being assigned correctly by 'divideind'

I am new in matlab and I am trying to use 'divideind' to divide my data in order, but unfortunately it gives me this:

trainInd: [1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 … ] (1×1418 double)

valInd: [1×0 double]

testInd: [1×0 double]

I would appreciate it if you tell me where is the problem?

this is my code:

X = data.inputs;

% Define the dependent variable
y = data.output;
Q = length(y);

% Define the split points
train_end = 1418; 
val_end = 1773;    
test_end = Q;      

% Assign the indices
train_indices = 1:train_end;
val_indices = (train_end + 1):val_end; 
test_indices = (val_end + 1):test_end; 

% Use divideind to ensure proper division
[train_indices,val_indices,test_indices] = divideind(Q,train_indices,val_indices,test_indices);

% Divide the data into training and testing sets
x_train = log(X(train_indices, :));
x_valid = log(X(val_indices, :));
x_test = log(X(test_indices, :));
y_train = log(y(train_indices));
y_valid = log(y(val_indices));
y_test = log(y(test_indices));

% Transpose the data to fit the input format of the neural network
inputs_train = x_train';
outputs_train = y_train';
inputs_validation = x_valid' ;
outputs_valid = y_valid';
inputs_test = x_test';
outputs_test = y_test';

% Initialize arrays to store RMSE values
train_rmse = zeros(1, 10);
Valid_rmse = zeros(1, 10);
test_rmse = zeros(1, 10);

for i = 1:10
    hiddenLayerSize = [i, i];
    net = fitnet(hiddenLayerSize, 'trainlm');
    net.trainParam.epochs = 200; 
    net.layers{1}.transferFcn = 'poslin'; 
    net.layers{2}.transferFcn = 'logsig'; 
    
    net.divideFcn = 'divideind';
    net.divideMode = 'sample';
    net.divideParam.trainInd = train_indices;
    net.divideParam.valInd = val_indices;
    net.divideParam.testInd = test_indices;
    
    [net,tr] = train(net, inputs_train, outputs_train);
    
    % Predict on training data
    train_predictions = exp(net(inputs_train(:,tr.trainInd)));
    yTrainTrue = exp(outputs_train(:,tr.trainInd));
    train_rmse(i) = sqrt(mean((train_predictions - yTrainTrue).^2));
    % Predict on Validation data
    Valid_predictions = exp(net(inputs_validation(:,tr.valInd)));
    yValTrue = exp(outputs_valid(:,tr.valInd));
    Valid_rmse(i) = sqrt(mean((Valid_predictions - yValTrue).^2));
    % Predict on Validation data
    test_predictions = exp(net(inputs_test(:,tr.testInd)));
    yTestTrue = exp(outputs_test(:,tr.testInd));
    test_rmse(i) = sqrt(mean((test_predictions - yTestTrue).^2));
    
end

Answer 1

[Steven Lord]

Can you confirm you're using the divideind function included as part of Deep Learning Toolbox?

which -all divideind

/MATLAB/toolbox/nnet/nnet/nndivision/divideind.m

When I run this code with a sample Q it gives me the results I'd expect:

callDivideInd(2000)

train_indices = 1x1418

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

val_indices = 1x355

1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

test_indices = 1x227

1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

Or perhaps you're calling it with a smaller Q, say Q equal to train_end?

callDivideInd(1418)

train_indices = 1x1418

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

val_indices =

  1x0 empty double row vector


test_indices =

  1x0 empty double row vector

function callDivideInd(Q)
train_end = 1418; 
val_end = 1773;    
test_end = Q;      
% Assign the indices
train_indices = 1:train_end;
val_indices = (train_end + 1):val_end; 
test_indices = (val_end + 1):test_end; 
% Use divideind to ensure proper division
[train_indices,val_indices,test_indices] = divideind(Q,train_indices,val_indices,test_indices)
end

[Fatemeh Choubineh]

Thank you so much for your answer. Yes I confirm that. The length of y is 2086 and I tried with different Q but still not working. I am stuck and I do not know what the problem is.

[Steven Lord]

Can you run that which command and show us exactly what it returns? And can you show us exactly what the variable named data looks like? We probably don't need to see the values, just the sizes of its fields should be sufficient.