Not all members of a NeuralFMU are re-defined when calling the NeuralFMU. That should be made consistent.

[juguma]

You can define a ME-NeuralFMU via https://github.com/ThummeTo/FMIFlux.jl/blob/77471fba7cd47648c7c6afdf5bab9f8050aaaf13/src/neural.jl#L87

You can call the created nfmu https://github.com/ThummeTo/FMIFlux.jl/blob/77471fba7cd47648c7c6afdf5bab9f8050aaaf13/src/neural.jl#L1044 with a series of arguments. Some of these arguments redefine the values of the member variables of the nfmu, some don't. This should be made consistent: Variables which redefine member values are (everything fine here):

• tSpan
• x_Start
• setup
• reset
• instantiate
• freeInstance
• terminate
• tolerance
• parameters <-fmu parameters
• p <-net parameters

Variables which exist as member values but are not overwritten:

• recordValues
• solver <-is only pushed as args to the actual solve
• saveat
• sensealg <- that is only a member of the fmu, maybe overwrite only with a warning!?

PS: haven't checked CS-NeuralFMU.