Icon

Auto SARIMA Playground

<p><strong>Auto SARIMA Playground</strong></p><p><br>A workspace designed to experiment with the <strong>Auto-SARIMA</strong> node for time series forecasting.<br>Explore and adjust the model parameters to find the best-fitting function for your data!</p><p>Make sure to download the data folder next to the workflow</p>

Auto Sarima Playground

Download the entire repository to fetch data with relative paths

Configure the seasonality and evaluate your model like in the first example

Default of the Auto-SARIMA: Seasonal Component Set to 0

When the seasonal component (S) in a SARIMA model is set to 0, it means the model does not account for any repeating seasonal patterns in the data — for example, weekly, monthly, or yearly cycles.

This simplifies the model to a standard ARIMA, which captures trend and autocorrelation effects, but not periodic ones.


Why High Seasonality Values with SARIMA Should Be Avoided

A high value of S (the seasonal period) indicates that the model is attempting to capture very long seasonal cycles. For example, setting S = 52 models yearly seasonality in weekly data (since there are 52 weeks in a year).

However, excessively large values of S can make the model computationally expensive:

  • The number of parameters grows significantly as S increases, leading to longer estimation times that will fill your memory.

  • Values larger than 100 should generally be avoided.

Model Configuration Notes
By default, the parameters are set to train an ARIMA model without a seasonal component.

If you enable seasonality, make sure that the seasonal period matches the recurring pattern observed in your data (for example, weekly, monthly, or yearly cycles). Refer to the node’s parameter descriptions for more details.

Here are some suggested starting points in case of seasonal model to balance speed and accuracy, it should work in most cases to find a good model.

(Same notation of the dialog is used to define the parameter)

  • Seasonal parameter (p:2-d:2-q:2)

  • Not seasonal parameters: (P:3-D:2-Q:2)

Note: Computing the logarithm of Y helps stabilize variance, reduce the impact of outliers, and make trends or relationships in the data more linear for modeling (it helps the estimation to converge)

Improve Optimization Time

Fitting a SARIMA model with high seasonality can be time consuming.

The default values are set relatively high. This makes the optimization process slower, but it increases the likelihood of converging to a well performing ARIMA model.

If you aim for faster optimization, consider reducing the optimization parameters of the Markov Chain that governs the simulated annealing algorithm.

Additionally, check the partial autocorrelation plot to identify the maximum relevant lags. Use these values to configure the Max AR and Max MA parameters in the node. Reducing these limits decreases the number of parameter combinations that need to be tested and can significantly speed up the process.

Temperature
CSV Reader
SensorsDownload the whole repoto read the data
CSV Reader
Sales of shampooShort series set (s=0)
CSV Reader
String to Number
Monthly beer production
CSV Reader
Model short term correlation only
(Partial) Autocorrelation Functions and Plots
Date&Time-based Row Filter
Try Auto ETS as well for a much faster fit.Make sure to set the seasonality.
ETS Learner
Do not set a high value of seasonality f.e. 365(more details in the notes)You can remove seasonality first before training the modeland train a simpler ARIMA To remove seasonality apply:-Fourier Transformation-Classical decomposition-Moving average
Auto-SARIMA Learner
String to Date&Time
Auto-SARIMA Learner
Series with longer seasonality
Line Plot
Auto-SARIMA Predictor
Auto-SARIMA Predictor
Sorter
Line Plot
Line Plot

Nodes

Extensions

Links