ReverseDiff.gradient incorrect with `if` and independent values. about reversediff.jl HOT 5 CLOSED

It turns out I was still using Julia v0.6.0-rc3 (the final release candidate just before the v0.6.0 release), just updated to Julia v0.6.0 and now I can reproduce it. AFAICT, that means it was JuliaLang/julia@e5e8be2 that introduced the bug here. I'll dig through that commit tomorrow. Thanks again for finding this!

Maybe it's something about my GPU? I don't think I have CUDA libraries b/c when I ran Pkg.update() last time Tensorflow complained about it:

ReverseDiff doesn't automatically run anything on the GPU; it falls back to however array operations are defined on your input array type.

from reversediff.jl.

I can't reproduce this...here's what I'm getting on my end (using ReverseDiff v0.1.4, the latest version):

julia> using ReverseDiff

julia> f(X) = sum([if x > 0; x else 0 end for x in X])
f (generic function with 1 method)

julia> ReverseDiff.gradient(f, [2,3,-1])
3-element Array{Int64,1}:
 1
 1
 0

I'll close this since it seems to be fixed on the latest ReverseDiff version, but please let me know if you encounter any issues with this.

from reversediff.jl.

Huh, that is super weird. I just installed it again on a new machine, and I have v0.1.4 as well. I also just ran Pkg.update():

$ julia
               _
   _       _ _(_)_     |  A fresh approach to technical computing
  (_)     | (_) (_)    |  Documentation: https://docs.julialang.org
   _ _   _| |_  __ _   |  Type "?help" for help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 0.6.0 (2017-06-19 13:05 UTC)
 _/ |\__'_|_|_|\__'_|  |  Official http://julialang.org/ release
|__/                   |  x86_64-apple-darwin13.4.0

julia> Pkg.update()
INFO: Updating METADATA...
INFO: Computing changes...
INFO: No packages to install, update or remove

julia> using ReverseDiff

julia> f(X) = sum([if x > 0; x else 0 end for x in X])
f (generic function with 1 method)

julia> ReverseDiff.gradient(f, [2,3,-1])
3-element Array{Int64,1}:
 0
 0
 0

julia>

Maybe it's something to do with my machine? I'm on a 2017 13" MacBook Pro, macOS 10.12.5.

Maybe it's something about my GPU? I don't think I have CUDA libraries b/c when I ran Pkg.update() last time Tensorflow complained about it:

INFO: Upgrading TensorFlow: v0.6.2 => v0.6.6
INFO: Building CUDAdrv
======================================================================[ ERROR: CUDAdrv ]======================================================================

LoadError: Could not find the CUDA driver library (specify the path to libcuda using the CUDA_DRIVER environment variable).
while loading /Users/nhdaly/.julia/v0.6/CUDAdrv/deps/build.jl, in expression starting on line 119

==============================================================================================================================================================

from reversediff.jl.

Also, for more context:

julia> Pkg.installed()["ReverseDiff"]
v"0.1.4"

julia> Pkg.installed()["ForwardDiff"]
v"0.4.2"

from reversediff.jl.

cool, glad to help! good luck!

from reversediff.jl.