DOC Add Predictive examples

[MarcoGorelli]

xref #1082 (comment)

I thought it might be clearer to slightly edit the existing examples in SVI and MCMCKernel, which already have runnable code, and then to refer to them in the Predictive kernel

Here's the part added to Predictive:

SVI:

MCMCKernel:

MCMC.run:

---

EDIT

some doctest failure...I'll fix this later, the example in numpyro/infer/util.py needs to just be code, but without doctest running on it

[fehiepsi]

Thanks, @MarcoGorelli! I think you can remove >>> to skip doctest.

[fehiepsi]

nit: rng_key1 -> rng_key

To make it clearer that we need to set data=None, I think you can provide a signature for model:

Given a model

    def model(X, y=None):
        ...
        return numpyro.sample("obs", likelihood, obs=y)

you can sample from the prior predictive:

    predictive = Predictive(model, num_samples=1000)
    y_pred = predictive(rng_key, X)["obs"]

If you also have posterior samples, you can sample from the posterior predictive:

    predictive = Predictive(model, posterior_samples=posterior_samples)
    y_pred = predictive(rng_key, X)["obs"]

[fehiepsi]

This code does not return any useful information unless you set data=None (which will give you posterior predictive samples). If you want to get posterior samples, then you can move those codes to the end and add a comment

>>> # get posterior samples
>>> predictive = Predictive(guide, params=params, num_samples=1000)
>>> samples = predictive(random.PRNGKey(1), data)

[fehiepsi]

I don't think we have a model here. I think you can add a small code snippet to get_samples method of MCMC.

[fehiepsi]

I think get_samples is a better place for this. Also maybe changing them to those samples?

[fehiepsi]

nit num_samples=1000

[MarcoGorelli]

Thanks for your feedback! I'll address your comments at the weekend (won't have time this week before then unfortunately)

[fehiepsi]

Please take your time. We can always defer doc improvements to later releases. :)

[fehiepsi]

I think you need an empty string after this to make lint pass.

[MarcoGorelli]

I think it might be the * - will try making it a r string

[MarcoGorelli]

OK, fixed, it just needed to be put into a code block, so with the previous line ending with :: rather than :

[MarcoGorelli]

now it looks like this:

[fehiepsi]

yeah, good catch!

[fehiepsi]

Awesome, thanks @MarcoGorelli!