Parameter inference is performed via Hamiltonian Monte Carlo (HMC) sampling using ten leapfrog steps per iteration with automatic step size adjustment according to Betancourt et al.32. Per default 20,000 iterations are performed with 5,000 iterations used as burn-in. The parameters \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\alpha }_{k},{\gamma }_{m,k}$$\end{document}αk,γm,k are randomly initiated by drawing from standard normal priors. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${t}_{m,k}$$\end{document}tm,k is always initialized with 0 to ensure unbiased model selection, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma }^{2}$$\end{document}σ2 is initialized with 1. If the data contains entries that are zero, a pseudocount of 0.5 is added to these zero counts to reduce numerical instabilities.
