Closed seabbs closed 1 week ago
Noted that #388 shifts to using accumlate_scan
to cast expected observation forward in time.
A search of the code base suggests that we are down to only Renewal
using scan
rather than accumulate_scan
. This is a fairly easy issue now.
Agree. We could either keep this for a large refactor of renewal processes (i.e to use #408) or just do a straight swap with fewer changes asap?
Agree. We could either keep this for a large refactor of renewal processes (i.e to use #408) or just do a straight swap with fewer changes asap?
I'm inclined towards a quick swap before the refactor:
In #369 we added new functionality for the
AR
latent model that introduce an abstract approach to extendingBase.accumulate
via step functions. This can be used to replace our current usage of thescan
function.This is likely to be attracted as it uses a base function which will likely be a major target of Auto-diff optimisations. It also makes it easier for us to introduce composability to our
scan
functions (i.e #385).