Feature engineering - Time Series to Supervised Learning.py

import numpy as np
import pandas as pd
from sklearn.preprocessing import StandardScaler


def preprocessing(df, n_feats, train_len, win_len, fh):
    """
    Preprocess to return training and test batches.

    Args:
        df (pd.DataFrame): The DataFrame that contains the data, with the target at the first colum.
        n_feats (int): Number of features.
        train_len (int): Training set length.
        win_len (int): Input window length.
        fh (int): Forecasting horizon.

    Returns:
        Flattened training/test features/targets in mini-batches
    """

    # Split dataset
    train, test = df.iloc[:train_len, :].values, df.iloc[train_len - win_len:, :].values
    test_len = len(test)

    # Normalization
    normalizer = StandardScaler()
    train = normalizer.fit_transform(train)
    test = normalizer.transform(test)

    # Mini-batching
    X_train_batches = np.zeros([train_len - win_len - fh, win_len + n_feats - 1])
    y_train_batches = np.zeros([train_len - win_len - fh, fh])
    for i in range(0, train_len - win_len - fh):
        X_train_batches[i] = np.concatenate((train[i:i + win_len - 1, 0], train[i + win_len - 1, :]), axis=None)
        y_train_batches[i] = train[i + win_len:i + win_len + fh, 0]

    X_test_batches = np.zeros([test_len - win_len - fh, win_len + n_feats - 1])
    y_test_batches = np.zeros([test_len - win_len - fh, fh])
    for i in range(0, test_len - win_len - fh):
        X_test_batches[i] = np.concatenate((test[i:i + win_len - 1, 0], test[i + win_len - 1, :]), axis=None)
        y_test_batches[i] = test[i + win_len:i + win_len + fh, 0]

    return X_train_batches, y_train_batches, X_test_batches, y_test_batches, normalizer