Forecasts generated for each series in a hierarchy often violate aggregation constraints. The packages bayesRecon (R) and bayesreconpy (Python) implement probabilistic forecast reconciliation via conditioning for hierarchical time series.

Our packages reconcile probabilistic forecasts to ensure coherence across the hierarchy. They support different types of base forecasts: Gaussian, continuous non-Gaussian, discrete. They also support mixed hierarchies with discrete and continuous forecasts on different levels. The interface is unified across both languages, and the packages are released under an LGPL (≥3) license.

This presentation will introduce the software and demonstrate core functionalities with both R and Python short tutorials.