Invalid training data Predictors must be a N-by-1 cell array of sequences

% Reshape training data for LSTM input and output

num_features = size(X_train, 2);

num_samples_per_day = 24; % Number of hours in a day

num_days_train = size(X_train, 1) / num_samples_per_day;

X_train_reshaped = cell(num_days_train, 1);

y_train_reshaped = cell(num_days_train, 1);

for a = 1:num_days_train

start_idx = (a - 1) * num_samples_per_day + 1;

end_idx = a * num_samples_per_day;

% Extract data for the current day

current_day_data = X_train(start_idx:end_idx, :);

current_day_target = y_train(start_idx:end_idx);

% Reshape data into a cell array of matrices [num_features x num_samples_per_day]

reshaped_data = num2cell(current_day_data', 1);

X_train_reshaped{a} = reshaped_data;

% Reshape target values into a cell array of column vectors

reshaped_target = reshape(current_day_target, [], 1);

y_train_reshaped{a} = reshaped_target;

end

% Reshape validation data for LSTM input

num_days_val = size(X_val, 1) / num_samples_per_day;

X_val_reshaped = cell(num_days_val, 1);

y_val_reshaped = cell(num_days_val, 1);

for b = 1:num_days_val

start_idx = (b - 1) * num_samples_per_day + 1;

end_idx = b * num_samples_per_day;

% Extract data for the current day

current_day_data = X_val(start_idx:end_idx, :);

current_day_target = y_val(start_idx:end_idx);

% Reshape data into a matrix [num_features x num_samples_per_day]

reshaped_data = current_day_data';

% Reshape target values into a column vector

reshaped_target = current_day_target';

% Store reshaped data and target for the current day

X_val_reshaped{b} = reshaped_data;

y_val_reshaped{b} = reshaped_target;

end

% Reshape test data for LSTM input

num_days_test = size(X_test, 1) / num_samples_per_day;

X_test_reshaped = cell(num_days_test, 1);

for c = 1:num_days_test

start_idx = (c - 1) * num_samples_per_day + 1;

end_idx = c * num_samples_per_day;

% Extract data for the current day

current_day_data = X_test(start_idx:end_idx, :);

% Reshape data into a matrix [num_features x num_samples_per_day]

reshaped_data = current_day_data';

% Store reshaped data for the current day

X_test_reshaped{c} = reshaped_data;

end

% Define hyperparameters for grid search

hidden_layers = [1, 2, 3];

num_hidden_units_range = [24, 48, 96, 192];

dropout_rates = [0.1, 0.2, 0.3, 0.4, 0.5];

learning_rates = [0.05, 0.01, 0.005, 0.001, 0.0005];

optimization_solvers_list = {'adam', 'sgdm', 'rmsprop'};

num_epochs_range = [50, 100, 150, 200, 250];

% Build LSTM model

layers = [

sequenceInputLayer(num_features)

];

for m = 1:num_layers

layers = [

layers

lstmLayer(num_units, 'OutputMode', 'sequence')

dropoutLayer(dropout_rate)

];

end

% Connect the last LSTM layer to a fully connected layer

layers = [

layers

fullyConnectedLayer(num_units)

dropoutLayer(dropout_rate)

fullyConnectedLayer(24) % Output layer with 24 units for 24-hour GHI forecasting

regressionLayer % Output layer for regression tasks

];

% Set options for training

options = trainingOptions(optimizer, ...

'MaxEpochs', num_epochs, ...

'MiniBatchSize', 64, ...

'InitialLearnRate', learning_rate, ...

'LearnRateSchedule', 'piecewise', ...

'LearnRateDropPeriod', 50, ...

'LearnRateDropFactor', 0.1, ...

'ValidationData', {X_val_reshaped, y_val_reshaped}, ...

'ValidationFrequency', 10, ...

'ValidationPatience', 20, ...

'Verbose', 0, ...

'Plots', 'training-progress');

% Train the LSTM model

lstm_model = trainNetwork(X_train_reshaped, y_train_reshaped, layers, options);

% Predict GHI using the trained model on validation data

y_pred_val = predict(lstm_model, X_val_reshaped);

% Denormalize predictions

y_pred_denormalized = y_pred_val * (max(features.GHI) - min(features.GHI)) + min(features.GHI);

% Calculate evaluation metrics

rmse_val = calculate_rmse(y_val, y_pred_val_denormalized);

mae_val = calculate_mae(y_val, y_pred_val_denormalized);

end

Error using trainNetwork (line 184)

Invalid training data. Predictors must be a N-by-1 cell array of sequences, where N is the number of sequences. All

sequences must have the same feature dimension and at least one time step.

[Ayush Aniket]

Hi Babajide,

From the error it seems that the input data for training the LSTM is not in the correct format. Can you share the data to reproduce the error?