How to define a custom AD rule in ForwardDiff using DiffRules. I have tried this: using DiffRules using ForwardDiff f(xx) = NaN DiffRules.@define_diffrule Main.f(xx) = :($xx[1] + $xx[2], $xx[1] - $xx[2]) julia> ForwardDiff.gradient(f, [1, 20]) 2-element Vector{Float64}: 0.0 0.0 as you can see…

You’ve preserved the tag which is good, the first type parameter, but I think you need to allow for multiple partials: julia> using ForwardDiff: Dual, partials, jacobian julia> g(x::Dual{Z,T,N}) where {Z,T,N} = Dual{Z}(NaN, ntuple(i -> 5 * partials(x,i), N)) g (generic function with 2 methods) ju…

How to define custom AD rule in ForwardDiff

General Usage

amrods August 27, 2021, 12:08am 6

I see. Going back to the first post, just to complete the learning round, how can I define custom rules with DiffRules that can be used with ForwardDiff?

Topic		Replies	Views
Issue with ForwardDiff custom AD rule Specific Domains question , forwarddiff	7	710	December 10, 2021
New differentiation rules for ForwardDiff General Usage differentiation	0	680	December 26, 2018
ChainRulesCore and ForwardDiff Numerics question	15	2477	October 21, 2022
Registering Analytical Derivative with ReverseDiff/ForwardDiff Optimization (Mathematical)	4	805	April 25, 2018
Nesting ForwardDiff inside ReverseDiff? General Usage question , differentiation	5	1308	January 4, 2018

How to define custom AD rule in ForwardDiff

Related topics