For calibrate_model, by default use random initial guess?

[TorkelE]

Would it make sense to, for calibrate_model, if an initial parameter guess (p0) is not provided, simply randomly select one from within the bounds (and priors) of the parameters? It feels like this will typically be a random value, in which case providing functionality for it might be good.

[sebapersson]

Hmm, I am against this, as I would like calibrate_model to mimic most available optimizers that all require a start-guess to be generated.

But it would be nice to to have functionality for start-guesses, so maybe a good solution would be to export the generate_startguesses function, and make it return a vector in case the user only wants 1 start-guess?

[TorkelE]

Consistency is probably best here, and a generate_startguesses should serve the same purpose as well.