If anyone is having some trouble understanding the semantics behind
event_shape or want to know more after reading Tensor shapes in Pyro I would recommend reading the excellent post Reasoning about Shapes and Probability Distributions by Eric J. Ma.
Another source would be TensorFlow’s probability post Understanding TensorFlow Distributions Shapes (where I found Eric Ma’s post). Just be careful because not everything is applicable to Pyro and NumPyro since it’s TensorFlow.
Hope it helps. =)