Jimmy Jia

Can we hard-memoize only the most recent value, and do it per schema? A little tricky where to store the per-field cached accessors anyway, no? I guess it could be...

The cost analysis case example is interesting, because in the context of using the programmatic API, I think it's suboptimal to set up costing via directives. Instead of the above,...

I'll also note that some of the concerns around metadata are specifically around communicating more information to clients. A canonical example here is augmenting type system information by informing the...

Looks like 1.1.0 is getting this as `LazyAdamOptimizer` with the sparse behavior as described? https://github.com/tensorflow/tensorflow/blob/v1.1.0-rc1/tensorflow/contrib/opt/python/training/lazy_adam_optimizer.py

It is debounced, no? We only do work if the rAF handle isn’t set. If you have a test setup, maybe try making it passive? It shouldn’t help, but maybe...

I mean, “not doing anything if the timer is already set” is pretty well debounced, no?

Okay, thinking about this a little more – a lot of the work in the rAF handler (which is causing the stutter) on window scroll is to work around quirks...

That sounds like a good idea to me. Ideally we should avoid checking/saving stuff while the user is actively scrolling as much as possible.

Care to take a stab at a PR?

`requestIdleCallback` would be good, if there were a good polyfill.