Hi. I want to know if Pyro supports defining model by log likelihood.
In Stan, there are two ways of modeling,
- directly specifying distribution:
y ~ P(data, param)e.g. 1.1 Linear regression | Stan User’s Guide
- accumulating log likelihood:
target += log_prob(data, param)e.g. 5.3 Summing out the responsibility parameter | Stan User’s Guide
The first method is equivalent to Pyro’s
pyro.sample. Does Pyro also support the second way?