Observations being read as number of columns instead of rows

Before splitting my data into training, validation, and testing, the code worked fine. After the split, trainnet begins reading the observations as the number of columns instead of by the number of rows. I checked the size of the array files and they are the same size the arrays were before I added a data split. I do not understand why trainnet starts reading the arrays differently. Any help would be appreciated.

Error message:

Error using trainnet

Error forming validation data mini-batch.

Error in MainCode6 (line 40)

netTrained = trainnet(stan_x_train, stan_y_train, net, "mse", opts);

Caused by:

Number of observations in predictors (6) and targets (1) must match. Check that the data and network are

consistent.

% Script Number One
all = readtable(excelFile,"Sheet","Nsplit6",'VariableNamingRule','preserve');

train_ratio = 0.8;  
val_ratio = 0.1;    
test_ratio = 0.1;   

num_samples = size(all, 1);

indices = randperm(num_samples);

num_train = floor(train_ratio * num_samples);
num_val = floor(val_ratio * num_samples);
num_test = num_samples - num_train - num_val;
train_indices = indices(1:num_train);
val_indices = indices(num_train+1:num_train+num_val);
test_indices = indices(num_train+num_val+1:end);

train_data = all(train_indices, :);
val_data = all(val_indices, :);
test_data = all(test_indices, :);

X_train = train_data(:, 1:end-1);  
Y_train = train_data(:, end);

X_val = val_data(:, 1:end-1);
Y_val = val_data(:, end);

X_test = test_data(:, 1:end-1);
Y_test = test_data(:, end);

stan_x_train = table2array(X_train);
stan_y_train = table2array(Y_train);
stan_x_val = table2array(X_val);
stan_y_val = table2array(Y_val);
stan_x_test = table2array(X_test);
stan_y_test = table2array(Y_test);


%Script number two
inputSize = 6; 
hiddenLayerSize1 = 40; 
hiddenLayerSize2 = 20;
hiddenLayerSize3 = 10;
outputSize = 1; 


layers = [
    featureInputLayer(inputSize)
    fullyConnectedLayer(hiddenLayerSize1, 'Name', 'fc1')
    reluLayer('Name', 'relu1')
    fullyConnectedLayer(hiddenLayerSize2, 'Name', 'fc2')
    reluLayer('Name', 'relu2')
    fullyConnectedLayer(hiddenLayerSize3, 'Name', 'fc3')
    reluLayer('Name', 'relu3')
    fullyConnectedLayer(outputSize, 'Name', 'output')
];

net = dlnetwork(layers);

opts = trainingOptions('adam', ... 
    'InitialLearnRate', 0.01, ...
    'MaxEpochs', 200, ... 
    'MiniBatchSize', 20, ...
    'ValidationData', {stan_x_val', stan_y_val'}, ... 
    'ValidationFrequency', 20, ... 
    'Verbose', true ...
    ); 


netTrained = trainnet(stan_x_train, stan_y_train, net, "mse", opts);

Answer 1

[Stephen23]

Perhaps because the data are (complex conjugate) transposed here:

'ValidationData', {stan_x_val', stan_y_val'},

Note that you can very easily simplify your code by using curly braces to access the table content:, e.g. replace:

X_train = train_data(:, 1:end-1);
..
stan_x_train = table2array(X_train);

with

stan_x_train = train_data{:,1:end-1};
%                        ^         ^ curly-braces = table content

See: https://www.mathworks.com/help/matlab/matlab_prog/access-data-in-a-table.html

[Candace]

Thank you, this helped. I am trying to accept your answer, however, the button is not visible.