Package: lopensemble 0.1.2.9000

lopensemble: Create Mixture Models From Predictive Samples

Combines predictions from individual time series or panel data models into an ensemble using stacking (Yao, Vehtari, Simpson, and Gelman (2018) <doi:10.1214/17-BA1091>) based on the Continuous Ranked Probability Score (CRPS) (Gneiting and Raftery (2007) <doi:10.1198/016214506000001437>) over k-step ahead predictions. Predictions must be predictive distributions represented by samples, typically posterior predictive simulation draws from a Markov chain Monte Carlo (MCMC) algorithm. Given training data with observed values and predictive samples from different models, optimal stacking weights are computed to minimize expected cross-validation predictive error. These weights can then be used to generate samples from the mixture model by drawing from individual model predictions in the correct proportions.

Authors:Nikos Bosse [aut, cre, cph], Yuling Yao [aut], Sam Abbott [aut], Sebastian Funk [aut]