Caching intermediate vectorized results with Aesara

[denehoffman]

I'm working on a project that fits columnar data with a model, and I think I want to use Aesara to create the models. Consider a model like z * myfunc(x, y), where myfunc is some computationally expensive graph. If my data provides both x and y, I want to just evaluate myfunc(x, y) and replace it with the values it gives. This would be applicable in cases where z is a free parameter, and in a fit I don't really need to calculate myfunc every single time I update my free parameters.

So my question, before I hack this in manually and possibly reinvent some wheels, is: is such caching behavior possible before compilation (without much additional work on my end)?

[denehoffman]

Apparently I don't read very comprehensively, because I just noticed the constant folding optimization which I believe does exactly this.