We present two extensions of the R package fable: fable.intermittent and fable.bayesRecon. The former brings probabilistic forecasting methods for count and intermittent time series into the fable ecosystem. The package also introduces the first R implementation of a Tweedie exponential smoothing model. The package fable.bayesRecon brings, for the first time, reconciliation via conditioning into the fable framework. Those methods, first introduced in the package bayesRecon, allow the reconciliation of continuous, mixed, and count forecast hierarchies. We demonstrate the core functionality of both packages and outline directions for future development.

We provide a quick overview of preference learning with Gaussian processes, and present a case study on bicycle route selection in the city of Paris. We show how to model user preferences over routes, and how to use the model to recommend routes that are likely to be preferred by a given user.

An overview of bayesRecon, an R package for probabilistic reconciliation of hierarchical time series forecasts via conditioning, covering the underlying reconciliation methods for continuous, mixed, and count forecast hierarchies.

A tutorial-style overview of preference learning with Gaussian processes, covering object preferences, label preferences, and learning from choice data, based on joint work with A. Benavoli, D. Piga, and D. Moranda.

Modeling user choice behavior with Gaussian process based choice functions, with an application to choice-based Bayesian optimization (ChoiceBO). Joint work with A. Benavoli and D. Piga.