I am doing a simple inference problem in which i have the following model:
with pm.Model() as model: prior_mu = pm.Normal('prior_mu', mu=mu) prior_std = pm.Normal('prior_std', mu=std) prior = pm.Normal('prior',prior_mu,prior_std) pm.Normal('Likelihood', mu=prior, observed=pieces)
where pieces is a signal that is sometimes turned off, and other times has a value quite far from zero. See figure below
The prior mu and sd are the expected value if the pieces-input was constant. After infering this, i get an extremely narrow posterior. Shouldnt the posterior distribution be rather wide as the observed values vary so much?
see figure of the inferred, extremely narrow posterior below.