[New Extension] How to do Time Series Forecasting with just 1 Operator!
Forecasting is targeted to make the forecasting of time series easy. It provides two operators, which forecast the next steps of a time series. The extension is designed to remove the complexity and ensuring good practices in hyper parameter tuning and validation.
The operator “Forecast Univariate” allows you to use statistical methods like ARIMA, Holt-Winters and Functional Seasonal Decomposition forecasting with one simple operator. A standard analysis of a data set looks like this:
As you see we only provide the data set to be forecasted and all the rest is done in this operator. The operator provides three results.
The forecast of the next n rows:
The performance of the operation validated using Sliding Window Validation:
And the back tested data, so that you can easily compare label and forecast.
The same interface and results are returned by the new operator “Forecast (Multivariate)”. Contrary to the univariate case it can use additional attributes as depended variables. This allows you to use for example the diesel price to forecast the gas price and vice versa. All the complexity of Feature Aggregation, Optimization of Hyperparameters and Validation is wrapped into the operator. You can just use it out of the box!
Currently the operator uses Decision Trees and Linear Models to forecast the future. Other machine learning models are added shortly.
This extension is based on the custom operators’ extension. Custom operators are operators which created by using other operators. You can easily turn any workflow into an operator and ship it to your colleagues. This also means you can investigate our new operators and see what we are doing!