import numpy as np # note: this function was inspired by # Brownlee, J. (2018, November 13). How to Develop LSTM Models for Time Series # Forecasting. Machine Learning Mastery. https://machinelearningmastery.com/how-to-develop-lstm-models-for-time-series-forecasting/ def split_sequences_multi_output(sequences, target, n_steps:int, n_steps_out:int): ''' Splits multivariate time series data into input-output sequences. This function creates sequences of input data (X) and corresponding target outputs for forecasting tasks where both target variables and features are predicted. It extracts `n_steps` time steps as input and predicts the next `n_steps_out` time steps for both the target variable and the feature sequences. ''' X, y_target, y_features = list(), list(), list() for i in range(len(sequences)): end_ix = i + n_steps # end of this input window out_end_ix = end_ix + n_steps_out # end of the target window (72 steps ahead) if out_end_ix > len(sequences): break seq_x = sequences[i:end_ix, :] X.append(seq_x) if target is not None: seq_q_target = target[end_ix:out_end_ix, 0] seq_q_features = sequences[end_ix:out_end_ix, :] y_target.append(seq_q_target) y_features.append(seq_q_features) return np.array(X), np.array(y_target), np.array(y_features)