Suppose I have this situation

```
def model(Y=None):
a = numpyro.sample('a', dist.Normal(0,1))
b = numpyro.sample('b, dist.Normal(0,1))
f = f(a,b) # f is a generic function
f2 = f2(f) # f2 is another generic function
sigma = numpyro.sample('sigma', dist.Normal(0,1))
numpyro.sample('f2', dist.Normal(f2, sigma), obs=Y)
```

Once I infer the parameters `a,b`

and `sigma`

I would like to compute the predictive posterior on `f`

. I know that to get the predictive posterior on `f2`

I can use the class Predict, but how can I do it with `f`

?