Bayesian models begin with one set of plausibilities assigned to each parameter; called prior plausibilities. These priors are then updated, as the model learns from the data, to produce posterior plausibilities. At each step as the model learns, the updated set of plausibilities becomes the new initial plausibilities for the next observation. In other words, the Bayesian model updates the prior distributions with their logical consequences: the posterior distribution. In this post, I use a linear model in
pymc3, using all of its latest capabilities, so as to provide a complete and state-of-the-art example of generalized linear modeling.