Move post_init check into eval for kernels which provide it.

[femtomc]

@dfm Sorry -- I SNAFU'd my other PR on accident with my poor local git abilities.

This is a continuation of #146 -- and moves the post_init check into evaluate for kernels which perform the check.

Re -- I'm not sure if my moving of assert ... until after the checks will mess up typechecking -- let me know if you end up having a moment.

[femtomc]

Bumping this -- in case a maintainer can run the approved workflows (which I don't have access to).

[dfm]

Whoops - sorry for being slow! I've had a crazy couple of weeks. Thanks for the ping.

[femtomc]

@dfm Ah, so it looks like I need to change some of the tests - where you explicitly check for the failures I've moved.

[femtomc]

Hi @dfm -- sorry for the delay on finishing this (research and all). How do you want to handle Product checks -- e.g. here:

    # Check for dimension issues when multiplied and evaluated.
    with pytest.raises(ValueError):
        k = jnp.ones(3) * kernels.Matern32(1.5)
        k.evaluate(v, v)

with this change, users are allowed to create k - but evaluation will throw an error.

And, just to remind ourselves why I motivated this change (it's been a few weeks) -- sometimes I want to return kernels out of vmap calls (e.g. imagine returning a trace from a probabilistic program where we also record what kernel was invoked), which means that they should evaluate just fine, but dimensional checks on construction will fail for this pattern.

[dfm]

@femtomc — Absolutely no worries! This looks like a good approach, the only change I would make would be to isolate the test, i.e.:

    # Check for dimension issues when multiplied and evaluated.
    k = jnp.ones(3) * kernels.Matern32(1.5)
    with pytest.raises(ValueError):
        k.evaluate(v, v)

to make it a bit clearer. Otherwise I'm happy with this!

[dfm]

The docs failure isn't related to this PR - can you rebase when you get a chance?

For the "news" failure, can you add a file called news/148.feature that briefly describes this PR? The text will appear in the release notes.

Finally: can you also add your info to the .zenodo.json file so that you can be included in the author list for the Zenodo release?

[femtomc]

Thank you! Hopefully all done.

[femtomc]

Oops -- I missed your comment about isolating the test. Will fix.

[dfm]

Thanks @femtomc!! 🚀