Abstract

A tour of parameter adaptation and tuning for gradient-based samplers in PyMC3, focussing on what is informed by theory, and what is informed by practice. Special attention is paid to how tuning in PyMC3 compares and contrasts to tuning in Stan.