JuliaDiff / JuliaDiff/ForwardDiff.jl

Perturbation Confusion II: The Tag Wars

Open
#313 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Julia
Stars
1k
Forks
160
PR merge metrics
No merged PRs in 30d

Description

@vtjnash was able to rewrite the classic perturbation confusion example into a form that thwarts ForwardDiff's tag generation:

julia> using ForwardDiff

julia> const D = ForwardDiff.derivative
derivative (generic function with 4 methods)

julia> struct Confuse
                whoami::Bool
                x
              end

julia> function (c::Confuse)(xy)
                if c.whoami
                  return xy * D(Confuse(false, xy), 1)
                else
                  return c.x + xy
                end
              end

julia> D(Confuse(true, 0), 1) # this should be 1
2

Essentially, since the function type is constant whether or not the function recursively differentiates itself, ForwardDiff generates the same tag for the nested call that it generates for the outer call.

So it seems that sufficiently unique tags cannot be generated from local type information alone...

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reproducing the Julia REPL example in the issue and trace how ForwardDiff generates tags for the outer and nested derivative calls. The work is done when this recursive case produces the expected derivative of 1 and the tag collision is covered by a regression test.

Written by the indexing model from the issue text.

Assessment

Tech stack
julia
Domain
backend
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
20/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.