Skip to content

On-premise users: click in-app to access the full platform documentation for your version of DataRobot.

Time series hierarchical reconciliation

Access this AI accelerator on GitHub

This AI Accelerator demonstrates how to reconcile (e.g., post-processing to sum appropriately) independent time series forecasts with a hierarchical structure. Reconciling, also known as making "coherent" forecasts, is often a requirement when submitting hierarchical forecasts to stakeholders. This notebook leverages the increasingly popular HierarchicalForecast python library to do the reconciliation on forecasts generated from DataRobot time series deployments. The steps demonstrated are as follows:

  1. Installing hierarchicalforecast
  2. Importing libraries
  3. Loading the example dataset
  4. Preparing training data for each hierarchy
  5. Building models for each level
  6. Deploying models for each level
  7. Making forecasts
  8. Preparing the forecasts
  9. Reconcile forecasts
  10. Comparing forecasts
  11. Conclusion

Note that steps 2-6 steps are purely for providing example time series deployments and forecasts from those deployments (in case you don't have any). If you already have a set of forecasts you want to reconcile, feel free to skip to step 7.

Updated May 20, 2024