Frames White
Frames White
> Why is AD.jacobian so central to AbstractDifferentiation.jl > Why isn't AbstractDifferentiation.jl built around two `primitives value_and_pullback_function` and `value_and_pushforward` Historical reasons based mainly on the original author have a strong...
That is a cute idea, and I do like the other benefits of a macro.
Discovered while looking into issues on https://github.com/FluxML/Zygote.jl/pull/1035
Yes, we should do this. It is better for errors to indicate something is wrong rather than a failure we should do it in this package, too slow to try...
We mostly just don't test them at all right now, because the implementation is trivial, and if it is wrong it is probably a bug in `ChainRulesCore.@non_differentiable`. Sometimes we test...
The only fully ChainRules compatible forward-mode AD that is working right now is Diffractor.jl which is yet to tag a release. It does have a very nice forwards mode AD...
It would be good to add this
It's an interesting idea. But I feel like it would be hard to get accuracy. Also really expensive at higher dimensions? Not sure. Testing both sides of discontinualities is I...
yes, in general FiniteDifferences.jl doesn't work well on Float32.