Can't sample posterior predictive, if model uses no covariates

numpyro
1

Hi. I am trying to simulate data using the numpyro implementation of the stochastic volatility model, described in the docs. However, when I try to get posterior predictive samples, it just returns observed values. For clarity, here’s the model function below:

def model(returns):
    step_size = numpyro.sample("sigma", dist.Exponential(50.0))
    s = numpyro.sample(
        "s", dist.GaussianRandomWalk(scale=step_size, num_steps=jnp.shape(returns)[0])
    )
    nu = numpyro.sample("nu", dist.Exponential(0.1))
    return numpyro.sample(
        "r", dist.StudentT(df=nu, loc=0.0, scale=jnp.exp(s)), obs=returns
    )

So, when I try to predict after fitting the model:

...
predictive = Predictive(model=model, posterior_samples=mcmc.get_samples())
samples_predictive = predictive(random.PRNGKey(42), returns)

I just get my observed returns back.

Am I doing something wrong? Is there a way to actually forward sample it? Thank you!

2

“r” in the trace should have your predicted values

What variable do you want to simulate?

Do you maybe want this

def model(returns=None):
  if returns:
   numpyro.sample("r", ..., obs=returns)
  else:
    ...
3

Hi @dirknbr. Appreciate your reply. Yes, I also raised this question on github and the guys there helped me out. Long story short, I needed to make the observed data a kwarg and set it to None.