Compute log-likelihood when I use MCMC sampling

I want to compute the log-likelihood in pyro’s MCMC sampling. I find the method in numPyro and use similar method in pyro’s but it doesn’t work.

nuts_kernel = NUTS(model)
mcmc = MCMC(nuts_kernel, num_samples=50, warmup_steps=50,num_chains=1)
mcmc.run(X1,X2,Y)
sample=mcmc.get_samples()
conditioned_model=poutine.condition(model,sample)
model_trace=pyro.poutine.trace(conditioned_model).get_trace(X1,X2,Y)
obs_node=model_trace.nodes["y_obs"]
loglikehood=obs_node['fn'].log_prob(obs_node['value'])

Hi @everli, we have similar code in ArviZ. I think you can adapt it for your problem.