Resumen de Scaleable distributional regression

Nikolaus Umlauf, Thorsten Simon

• Estimation of distributional regression models using datasets beyond 106 observations is a dicult task. We propose a novel optimizer which is based on the ideas of stochastic gradient descent and can easily deal with large data sets.

  Moreover, the algorithm performs automatic variable and smoothing parameter selection and its performance is in most cases superior or at least equal to other implementations for distributional regression. An implementation is provided in the R package bamlss. We illustrate the usefulness of the approach by implementing a state-of-the-art prediction model for lightning occurrence and counts in complex terrain.