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
At the moment, calls in the forward mode work correctly only if called function has one input.
If a function f is called with args a1, a2, ..., its derivative should be a dot product of gradient(f)(a1, a2, ...) and a vector of derivatives of args: { da1, da2, ...}.
This dot product can be efficiently computed by differentiating f in the forward mode and providing values of da1, da2, ... as seeds. See issue #100.
The text was updated successfully, but these errors were encountered:
At the moment, calls in the forward mode work correctly only if called function has one input.
If a function
f
is called with argsa1, a2, ...
, its derivative should be a dot product ofgradient(f)(a1, a2, ...)
and a vector of derivatives of args:{ da1, da2, ...}
.This dot product can be efficiently computed by differentiating
f
in the forward mode and providing values ofda1, da2, ...
as seeds. See issue #100.The text was updated successfully, but these errors were encountered: