Holt Winter Experiment

data-realistic-2023-07-01-to-2023-11-30-dec.csv

main.py

# https://medium.com/analytics-vidhya/python-code-on-holt-winters-forecasting-3843808a9873
import datetime

# dataframe opertations - pandas
import pandas as pd
import matplotlib as mpl
from matplotlib import pyplot as plt
mpl.use('macosx')  # disables the plot a image and makes interactive

# Seasonality decomposition
from statsmodels.tsa.seasonal import seasonal_decompose # holt winters

# single exponential smoothing
from statsmodels.tsa.holtwinters import SimpleExpSmoothing   # double and triple exponential smoothing
from statsmodels.tsa.holtwinters import ExponentialSmoothing

def get_data():
    df = pd.read_csv('data-realistic-2023-07-01-to-2023-11-30-dec.csv')  # index_col='date_key', parse_dates=True, date_format='%Y-%m-%d %H-%M-%S'

    # Convert 'date_key' column to datetime format
    df['date_key'] = pd.to_datetime(df['date_key'])
    # Set 'date_key' as the index
    df = df.set_index('date_key')
    # Generate a date range with hourly frequency
    date_range = pd.date_range(start=df.index.min(), end=df.index.max(), freq='H')
    # Reindex the DataFrame with the complete date range and fill missing values with 0
    df = df.reindex(date_range, fill_value=0)

    return df

def explore():
    print("running")
    df = get_data()

    print(df.shape)  # having a look at the data
    print(df.head(24))  # plotting the original data
    df[['views']].plot()

    df.index.freq = 'H'
    decompose_result = seasonal_decompose(df[['views']][(24*7*8):(24*7*10)], model='add')
    decompose_result.plot()

    # Set the frequency of the date time index as Monthly start as indicated by the data
    # airline.index.freq = 'H'  # Set the value of Alpha and define m (Time Period)

    df['HWES3_ADD'] = (ExponentialSmoothing(df['views'], trend='add', seasonal='add', seasonal_periods=24)
                       .fit()
                       .fittedvalues)
    df[['views','HWES3_ADD']].plot(title='Holt Winters Triple Exponential Smoothing: Additive and Multiplicative Trend',
                                   xlim=[datetime.date(2023, 9, 1), datetime.date(2023, 9, 7)],
                                   ylim=[0, 30],
                                   figsize=(9, 6))
    plt.show()

def main():

    # explore()

    df = get_data()

    # Split into train and test set

    train_length = round(df.shape[0] * 0.7)

    train = df[:train_length]
    test = df[train_length:]

    fitted_model = ExponentialSmoothing(train, trend='add', seasonal='add', seasonal_periods=24).fit()

    test_predictions = fitted_model.forecast(24*7)
    train['views'].plot(legend=True, label='TRAIN')
    test['views'].plot(legend=True, label='TEST',figsize=(6, 4))
    test_predictions.plot(legend=True, label='PREDICTION')
    plt.title('Train, Test and Predicted Test using Holt Winters')

    # enable interactiv mode top for zoom

    # not recalculating HWS for every value like do in JS, which is more correct
    # these predictions looks technically correct and not bad if you only take the first 2 at a time


    plt.show()

if __name__ == '__main__':
    main()