time_series_params: Tuning Parameters for Time Series (ts-class) Models

time_series_paramsR Documentation

Tuning Parameters for Time Series (ts-class) Models


Tuning Parameters for Time Series (ts-class) Models


seasonal_period(values = c("none", "daily", "weekly", "yearly"))



A time-based phrase


Time series models (e.g. Arima() and ets()) use stats::ts() or forecast::msts() to apply seasonality. We can do the same process using the following general time series parameter:

  • period: The periodic nature of the seasonality.

It's usually best practice to not tune this parameter, but rather set to obvious values based on the seasonality of the data:

  • Daily Seasonality: Often used with hourly data (e.g. 24 hourly timestamps per day)

  • Weekly Seasonality: Often used with daily data (e.g. 7 daily timestamps per week)

  • Yearly Seasonalty: Often used with weekly, monthly, and quarterly data (e.g. 12 monthly observations per year).

However, in the event that users want to experiment with period tuning, you can do so with seasonal_period().



modeltime documentation built on June 8, 2022, 1:07 a.m.