The parameter fitting example from the documentation throws an error

[PBrdng]

Hi all,

I'm trying to run the parameter fitting example from the documentation. The line

optsol = solve(optprob, NLopt.LN_NELDERMEAD())

throws an error, namely

ArgumentError: invalid index: ModelingToolkit.ParameterIndex{SciMLStructures.Tunable, Int64}(SciMLStructures.Tunable(), 1, false) of type ModelingToolkit.ParameterIndex{SciMLStructures.Tunable, Int64}
Stacktrace:
  [1] to_index(i::ModelingToolkit.ParameterIndex{SciMLStructures.Tunable, Int64})
...

Any ideas what is the problem here? I updated all packages and Julia today, so I'm running the most recent version.

Thanks for your help!

[PBrdng]

Update: on the master branch everything works fine.

[isaacsas]

@TorkelE can hopefully help as he has some experience with DiffEqParamEstim, but my impression is that package really isn't being super maintained anymore.

Do you have issues if you directly use Optimization.jl and build your own custom loss function instead?

For the master version of docs we have already removed DiffEqParamEstimation from the tutorials, and now suggest just directly using Optimization.jl or PEtab.jl. The updated version of the tutorial you were looking at is:

https://docs.sciml.ai/Catalyst/dev/inverse_problems/optimization_ode_param_fitting/

[isaacsas]

(And at least for me, the updated tutorial works with the latest release version too.)

[PBrdng]

Thanks. Maybe, it's an issue with my setup. Nevermind, I will look into Petab. That's also what Torkel recommended to me.

[isaacsas]

Please do let us know if you try the newer tutorial version and it doesn't work for you too! Purely Optimization.jl workflows should work and are intended to be supported. It is just that DiffEqParamEstim.jl really isn't being maintained much (it's last release is over a year ago, for example).

All that said, my impression too is that if your use case fits within the functionality it provides, PEtab.jl is probably the best option for parameter inference in ODE reaction network models.

[isaacsas]

Here is the latest version of the PEtab.jl tutorial:

https://docs.sciml.ai/Catalyst/dev/inverse_problems/petab_ode_param_fitting/

[TorkelE]

Yes, just follow the latest PEtab tutorial, and that should hopefully have what you got. Unfortunately there has been a chain of MTK updates that have been stopping any Catalyst updates, so the docs are a bit outdated. Hopefully it is sorted out soon.

[ChrisRackauckas]

Yeah I need to update DiffEqParamEstim...