How to recover original variables after a reparameterization?

Misc.
1

Hi, I have a model that involves variables with projected normal and normal priors, e.g.,

def model():
   q = pyro.sample("q", dist.ProjectedNormal(torch.zeros(4)))
   t = pyro.sample("t", dist.Normal(torch.zeros(3), torch.ones(3)))
   ...

I parameterize the model using AutoReparam and run MCMC - everything is fine!

reparam_model = AutoReparam()(model)
kernel = RandomWalkKernel(reparam_model)
mcmc = MCMC(kernel)
mcmc.run(*args, **kwargs)
samples = mcmc.get_samples()

In samples, I only get the reparameterized variables, e.g. q_normal_decentered and t_decenteted, but I want to transform back to the original variables (as defined the model before reparameterization).

Is there a way to do so automatically?