JuliusMartensen

Hi! Sorry for the delay, but here is what I would do: ```Julia using DataDrivenDiffEq using OrdinaryDiffEq using DataDrivenSparse using Random using Plots # Generate the data function T(t) 10*(1+sin(t/200)*0.4+0.4)...

Note that the closure can use any stuff here. This is basically generating a similar expression to what you did above in the original system, but written as a chain...

I think this is not 100% what you have in mind, but it should work roughly. ```julia using SparseArrays using Symbolics a = sprand(10, 10, 0.1) nzrows, nzcols, nzvals =...

Yes! I'll work on that. I just ported the necessary stuff right now, but this is a good point. The missing documentation snippet can be found [here](https://datadriven.sciml.ai/v0.8/examples/3_linear_continuous_system_controls/), in the previous...

The `DirectDataDriven` is indeed just a wrapper for the functions above and can be found [here](https://github.com/SciML/DataDrivenDiffEq.jl/blob/b97ba99f1c3d35ea10bd677cbe6c619aee59e259/src/problem/type.jl#L256). *Edit* And within the [docs](https://docs.sciml.ai/DataDrivenDiffEq/stable/problems/#Concrete-Types). The regression / usage within the example is on...

I've added another simpler MWE. When using Zygote, the gradients match, which could indicate numerical inaccuracy.

Incrementally downgrading MTK to 9.80 did not help (didn't account for patch releases, though ). Same for SciMLSensitivity to 7.79 .

ReverseDiff also returns zeros.