Such a generic approach will allow the Relations to fit any reasonable workflow, because possible relations are defined by users, just like the Labels are.

the 2 hardest problems in computer science are essentially the 2 hardest problems of life in general, as far as humans and information are concerned.

The non-determinism is why cache invalidation — and that other hard problem, naming things — are uniquely and intractably hard problems in computer science. Computers can perfectly solve deterministic problems. But they can’t predict when to invalidate a cache because, ultimately, we, the humans who design and build computational processes, can’t agree on when a cache needs to be invalidated.

I like the answers already given, but I think they both could use an even more top-level, generalized explanation.

Cache invalidation is hard because: Everything in life we want to know, changes.Those changes are non-deterministic.

You can also fit generalized additive models (Chapter @ref(polynomial-and-spline-regression)), when linearity of the predictor cannot be assumed. This can be done using the mgcv package: