You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Many repos have a roadmap to v1.
This plan goes a little further.
The idea is:
ChainRules/ChainRulesCore v1:
It pragmatically works.
It is completely usable, and is in use by multiple AutoDiff systems by the time we tag v1.
But it might play fast and lose with the math, like it will probably keep something like the current defintion of AbstractDifferential, which conflates the idea of scaling and adding.
It has some edge cases that we just bail out on and don't handle representing. This may include complex deriviatives, and it probably will include mutation/nonpure functions.
It will support Julia 1.0
ChainRules/ChainRulesCore v2:
The math will be tighter
After a while of use in public we will have identified the actual edge cases that matter of things we can't represent. and it will be enhanced to support those
It may not support julia 1.0, e.g. might depend on features added in 1.3
This roadmap needs some refining,
but this was the big picture @willtebbutt and I were talking about
The text was updated successfully, but these errors were encountered:
Many repos have a roadmap to v1.
This plan goes a little further.
The idea is:
ChainRules/ChainRulesCore v1:
AbstractDifferential
, which conflates the idea of scaling and adding.ChainRules/ChainRulesCore v2:
This roadmap needs some refining,
but this was the big picture @willtebbutt and I were talking about
The text was updated successfully, but these errors were encountered: