Forecast future values

Use this recipe to use trained forecasting models to predict future values after your historical dataset.

Input Data

Trained model folder

Folder containing models saved by the Train and evaluate forecasting models recipe.

Dataset with future values of external features (optional)

Only required if you specified external features in the Train and evaluate forecasting models recipe. This should contain the same external features and time columns (and time series identifiers columns if Long format was used) used during training.

Output Data

Forecast dataset

Dataset with predicted future values and confidence intervals. Use it to build charts to visually inspect the forecast results.


Model Selection

Selection mode

Choose how to select the model used for prediction:

  • Automatic to select the best-performing model from the last training session

    • Performance metric: metric used to retrieve which model has performed best. Lower is better.

  • Manual to select yourself a training session and a trained model:

    • Training session: UTC Timestamp of the training session from which to retrieve a trained model. Choose “Latest available” to always select the last training session (useful for operationalization).

    • Model name: Name of the trained model to retrieve from the selected session.


Confidence interval (%)

Lower and upper bounds forecasts of the selected confidence interval will be computed.

Include history

If you want to keep historical data used in training in addition to future values in the output dataset

  • Sampling method: if you want to include only the last records of the historical time series (the N most recent data).

  • Nb. records: how many records to keep for each time series