More restrictive use of the rrule name

[yebai]

At the moment, we are using rrule in a generic way, in the sense that they can mean both actual rrule for a function and the generated gradient function for a tape. It is better to keep a distinction between these two for better clarity.

[willtebbutt]

At the moment, we are using rrule in a generic way, in the sense that they can mean both actual rrule for a function and the generated gradient function for a tape. It is better to keep a distinction between these two for better clarity.

I know we discussed this yesterday, but I'm now realising that I'm not entirely sure what you mean. Could you please elaborate?

[yebai]

It's about avoiding referring to distribution log density tests as part of rrule tests, and avoiding referring to generated gradient function for unrolled_funciton as rrule.

Technically, this is correct but can cause confusion. It is slightly abusing the rrule terminology I think.

[willtebbutt]

Ah, right. I agree regarding the location of the distribution tests.

I don't know that I follow your reasoning regarding the unrolled_function rrule!!: it is most certainly an rrule!!, so it ought to be referred to as an rrule!!. It's possible that you find it confusing because the documentation is still being developed, but I don't feel that it's an abuse of terminology. On the contrary, it is a completely correct use of terminology, and IMO it would be confusing to introduce another function purely for the sake of it.

[willtebbutt]

Resolved by #34