feature/prior array

[rhayes777]

Adds the Array class which functions as a PriorModel that creates numpy arrays of floats.

An array is defined by its shape and a prior.

array = af.Array(
    shape=(2, 2),
    prior=af.GaussianPrior(mean=0.0, sigma=1.0),
)

The prior is copied to each index meaning that a 2x2 array has four independent priors.

print(array.prior_count)
> 4

Arrays can be accessed and modified using indexing.

array[0, 0] = 1.0
print(array.prior_count)
> 3

They can be instantiated just as any other model class

instance = array.instance_from_prior_medians()
print(instance)
> [
     [1.0, 0.0],
     [0.0, 0.0],
  ]

[Jammy2211]

Don't merge yet -- I'll build a test case this week and see whether any issues come up.

Couple of things to think about, albeit best not to worry about them until we've done some testing:

1. Does this all play nicely with our JAX / Pytrees wrapping?
2. How does this play with search chaining API?
3. Is there a clean interface with the default-prior yaml configs?

[rhayes777]

Need to check JAX and search chaining. We mostly do config on the type -> prior relationship so it was hard to see how to carry this over

[Jammy2211]

I would make it so that the config default gives the same prior to everything on the numpy array.

We can worry about JAX and search chaining later on then!

[rhayes777]

I would make it so that the config default gives the same prior to everything on the numpy array.

We can worry about JAX and search chaining later on then!

So at the moment you define one prior and that gets copied to every slot. Would we want a config to determine the default for all Arrays globally?

[rhayes777]

Should now work with prior passing and JAX

[VascoSch92]

can you describe what happens if the parameter prior is not provided?

[VascoSch92]

you are converting self.indices to a list but you are usint the typing Tuple[int, ...] everywhere.