Adjoint through getindex(::NonlinearSolution, sym)
#952
Conversation
|
Rebase? |
|
@AayushSabharwal the failures don't seem related to the changes in this PR afaict. |
|
@ChrisRackauckas does this need rebasing again? I noticed some failures on the tests I had added, but that has been resolved since |
|
This does need a rebase. MTK and Catalyst downstream should pass. We should also split the MTK workflow, since otherwise it will take ~3 hours |
|
isn't https://github.com/SciML/SciMLBase.jl/actions/runs/14102757384/job/39502400404?pr=952#step:6:1683 supposed to be solvd by it? |
|
This PR is for getting the adjoints through |
|
Test failure on Zygote is FluxML/Zygote.jl#1564 |
| # @testset "AD Observable Functions for Initialization" begin | ||
| # iprob = prob.f.initialization_data.initalizeprob | ||
| # isol = solve(iprob) | ||
| # gs, = gradient(isol) do isol | ||
| # isol[w] | ||
| # end | ||
|
|
||
| # end |
There was a problem hiding this comment.
I assume this will be enabled later?
There was a problem hiding this comment.
Yes, this requires SciML/SciMLSensitivity.jl#1168
3 participants
Checklist
contributor guidelines, in particular the SciML Style Guide and
COLPRAC.
Additional context
The current dispatches do not work for the NonlinearSolution type. Needed for getting initialization working since all the unknowns and parameters in an initialization problem are converted to observables internally.