Added WeightedCategorical-Prior

[JasperMartins]

This pull request adds a WeightedCategorical-prior that subsumes the existing Categorical-Prior. Functionally, the only difference between the two is that WeightedCategorical supports the setting of relative weights for all categories.

[mj-will]

Hi @JasperMartins, thank for implementing this and apologies for the delay in looking over it.

The changes look good to me but following a recent change to the Categorical prior (see #947) I think this will need slightly reworking to address the conflicts. Do you think you'd have time/capacity to do this?

[mj-will]

LGTM, just one question from my side.

[ColmTalbot]

I agree, this is generally in good shape, just a few small comments from me.

[ColmTalbot]

I was testing this out today (it happened to be relevant to something I'm working on) and generally it looks good. I noticed that the new priors (and also the categorical prior currently) doesn't store some of the input arguments, this means that the __repr__ method fails and so it can't be used during sampling.

tl;dr: the __init__ methods need to set all input arguments in the class, e.g.,

    ....
    self.ncategories = ncategories
    ....

[ColmTalbot]

Hi @JasperMartins just a gentle ping to see if you still have bandwidth to work on this change?

[JasperMartins]

Hi @ColmTalbot,

Sorry for the delay, I had been on vacation.
So, prompted by the init-from-repr issue, I have added the new priors to the general assembly of prior-tests, to catch other issues, which also uncovered a small bug in the cdf function for values below the minimum value in the prior-support.
I have fixed these issues and brought the branch up to date.

In principle, the tests in the file analytical_test.py could now probably be dropped.

[ColmTalbot]

Thanks for making these changes! No worries about the delay. I'm happy with the small duplication in the tests.

[mj-will]

LGTM, thanks for your work on this!