General information

Prepare Future Dataset brick generates a future time series dataset for a trained Neural Prophet model.

Description

Brick Locations

Bricks → Machine Learning → Prepare Future Dataset

Brick Parameters

• Number of periods

  The number of steps to extend the dataset into the future.

  <aside> ❗ Note that this value is ignored when the auto-regression components are present in the trained input model. In this case, the number of periods is defined by the number of forecasts in the model.

  </aside>

• Number of historic predictions

  The number of historic observances to include.

Future Regressors

• Datetime (optional)

  A column from the optional future regressors input that contains the dates of the future observations. If not selected, the consecutive dates are generated based on the model’s training data.

Events

• Datetime

  A column from the optional events input that contains the dates for the user-specific events for the forecast period.

• Events column

  A column from the optional events input that specifies the names of the user-specific events.

Brick Inputs/Outputs

• Inputs

  • data

    Dataset to be extended into the future. If the auto-regression options are defined for the model or the Number of Historic predictions > 0, the input must contain all columns used for model fit. The columns that did not appear in the model configuration will be automatically filtered.

  • model

    A trained Neural Prophet model.

  • optional future regressors

    This output is required if the input model was trained with future regressors. A future regressors dataframe must contain one column for each of the future regressors and optionally a datetime column if it is selected in the settings. All the other columns will be automatically filtered.

  • optional events

    Future event occurrences corresponding to ’periods’ steps into the future. It must contain columns ‘Datetime’ and ‘Events column’, selected in the settings. If the events were specified for the model but this input is not connected, then the custom events will be treated as non-occurring in the future.

• Outputs

  Brick returns a future dataset with new timestamps added to the historic data.

Example of usage

Here are examples of the Prepare Future Dataset brick usage.

First, let’s see how to apply the brick to the simple trained model without autoregressive components.