SARIMA
ARIMA extended with seasonal components for periodic time series
SARIMA (Seasonal ARIMA) extends ARIMA with seasonal AR, differencing, and MA terms. It captures both trend-based and repeating seasonal patterns in univariate time series.
When to use:
- Time series with clear seasonal patterns (weekly, monthly, yearly cycles)
- Sales, energy consumption, weather, or web traffic data
- Univariate forecasting with known seasonality period
Input:
- Trained model checkpoint — exported SARIMA fit
- Preprocessing config — scaling settings
- Training tail — last N observations for lag features
- Steps — forecast horizon in time steps
Output: Forecasted values for the specified horizon
Model Settings (set during training, used at inference)
AR Order (p) (default: 1) Non-seasonal autoregressive lag order.
Differencing (d) (default: 0) Non-seasonal differencing order.
MA Order (q) (default: 0) Non-seasonal moving average order.
Seasonal AR (P) (default: 1) Seasonal autoregressive order.
Seasonal Differencing (D) (default: 1) Seasonal differencing order.
Seasonal MA (Q) (default: 1) Seasonal moving average order.
Seasonal Period (m) (default: 12) Number of time steps per season (e.g., 12 for monthly data with yearly seasonality).
Inference Settings
No dedicated inference-time settings.