Abstract

Stan is a probabilistic programming language for specifying statistical models. Stan provides full Bayesian inference for continuous-variable models through Markov Chain Monte Carlo methods such as the No-U-Turn sampler, an adaptive form of Hamiltonian Monte Carlo sampling. Penalized maximum likelihood estimates are calculated using optimization methods such as the limited memory Broyden-Fletcher-Goldfarb-Shanno algorithm. Stan can be called through R using the rstan package, and through Python using the pystan package. Both interfaces support sampling and optimization-based inference with diagnostics and posterior analysis. In this talk it is shown a brief glance about the main properties of Stan. It is shown, also a couple of examples: first one related with a simple Bernoulli model and second one, about a Lotka-Volterra model based on ordinary differential equations.