Help Topics Concepts Package Class
The Time Series Filter Bean panel provides these options:
A Window and Horizon setting of 1 and 0 means the filter does not manipulate the input stream.
The Time Series Filter Bean panel is used to collect specific records from a datasource and place them in its outputbuffer. It is used in the Neural Prediction Agent when a window or horizon is specified.
Consider that each record in the data source represents one period's values. The Horizon value is the number of periods from the current record to the period to be predicted. If more than one period's values are to be used in a prediction, set the Window to the number of periods to use as the input. This allows a future value to be projected from several periods' worth of data. Some illustrations:
Window | Horizon | Description |
1 | 0 | Data for period 1 is used to predict the target value provided in record 1. |
2 | 0 | Data for periods 1 and 2 is used to predict the target value in record 2. |
2 | 1 | Data for periods 1 and 2 is used to predict the target value in record 3. |
For example, if the data source contains 52 weeks of data and data is available bi-weekly, the prediction may include 4 periods of data (the Window size) and the projection may be 2 weeks ahead (the Horizon value). Since an import source is a continuous loop, the values for weeks 49, 50, 51, and 52 would be used to predict the target supplied in week 2.
Note that changing the Window value changes the number of outputs from the filter. Hence the inputBuffer of any connected object will change in size according. The number of inputs values for neural network learning beans may need to be reset manually.