Holt's Winter Exponential Smoothing - Heisenberg0203/GETS GitHub Wiki

Holts’ Winter Exponential Smoothing (HWES) or Triple Exponential Smoothing (TES) is a robust model which captures level and trend along with seasonality component.

Forecast equation for HWES:


level_t = level_t-1 + trend_t-1 + (alpha) * (observed_t-1 - level_t-1 - trend_t-1 - season_t-m)

trend_t = trend_t-1 + beta *(level_t - level_t-1 - trend_t-1 )


season_t = season_t-m + gamma*(observed_t-1 -level_t-1 - season_t-m)

forecast_t+k = level_t + trend_t

Grammar for HES:

As discussed in HES (https://github.com/Heisenberg0203/GETS/wiki/Holt's-Exponential-Smoothing) alpha and beta controls the smoothing of level and trend respectively while gamma is the smoothing factor for seasonality over a period of time. This requires evolving five parameters, mainly alpha, beta, gamma, period_of_seasonality and step which are substituted in the above equation to make forecast using HWES.

Sample Derivation Tree:

Executing HES ( Please make sure that you are in src folder)

python ponyge.py --parameters gets/tes.txt

By default it runs on Daily Waste Generated dataset. Command to run on the custom dataset:

python ponyge.py --parameters gets/tes.txt --dataset_train path_to_traindataset --dataset_test path_to_testdataset.