Holt's Exponential Smoothing - Heisenberg0203/GETS GitHub Wiki

Holts’ Exponential Smoothing (HES) or Double Exponential Smoothing extends the concept of Simple Exponential Smoothing (SES) to capture trend along with level in the time series data to predict future values assuming there is no seasonality component in series

Forecast equation for SES:


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

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


forecast_t+k = level_t + trend_t

Grammar for HES:

The level is the forecast of the value in the series, which is controlled by the parameter alpha. While the trend is expected growth in the series, controlled by another parameter beta, the smoothing constant for trend. The generated optimal parameters for alpha, beta and step are derived through alpha_var, beta_var and step_var which is then substituted in the equation for forecasting using HES. In HES, the value of _**gamma **_is substituted as zero, since it assumes there is no seasonality.

Sample Derivation Tree:

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

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

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

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