Autodiff of solver clashes with Turing

[sciml]

Autodiff of solver clashes with Turing

[mohamed82008]

This is working for me on #127.

[mohamed82008]

Actually it's not, sorry I confused this issue with another.

[jonasmac16]

Do you know where exactly the issue lies causing this error?

[mohamed82008]

It seems like a ForwardDiff bug. I would find the exact function call that errors and exact numbers then open an issue there.

[ChrisRackauckas]

Probably need to awaken the great @jrevels . The issue likely comes from the fact that in SparseDiffTools we have to define our own Dual tags since we need to build the color vectors.

https://github.com/JuliaDiff/SparseDiffTools.jl/blob/master/src/differentiation/compute_jacobian_ad.jl#L46

[jonasmac16]

As suggested, I have opened an issue at ForwardDiff ( JuliaDiff/ForwardDiff.jl#446 ). Is there anything I can do to help to resolve this?

[ChrisRackauckas]

@jrevels James is in town tomorrow: you want to join our dinner at Pepper Sky and discuss this a bit?

[jonasmac16]

Hi, has there been any progress on this? Can I help somehow?

[ChrisRackauckas]

Fixed by JuliaDiff/SparseDiffTools.jl#105 and #157 will be a regression test for this.