Add option in Dud for max total iterations

[erikpelgrim]

Currently in Dud there is only the possibility to limit the number of outer iterations and inner iterations (within an outer iteration).

In order to have full control of the total iterations (total number of model runs) there will be more control on the total amount of run time.

This feature is requested by users calibrating the river bed roughness of the full Rijn in the Netherlands.

[erikpelgrim]

Also check why there should not be stopped on an inner iteration:

Unclear why these lines of code have been added in BaseDudCoreOptimizer.isThereMoreToDo():

	// do not stop on a backtracking iteration
	if (innerIter>0 && (imain < maxit)) moreToDo = true;

[nilsvanvelzen]

Also check why there should not be stopped on an inner iteration:

Unclear why these lines of code have been added in BaseDudCoreOptimizer.isThereMoreToDo():

	// do not stop on a backtracking iteration
	if (innerIter>0 && (imain < maxit)) moreToDo = true;

I would have to look in the code but I do not rule out that the last value is "rubbish" at least not the best approximation when it is in the backtrack modes.

[nilsvanvelzen]

Currently in Dud there is only the possibility to limit the number of outer iterations and inner iterations (within an outer iteration).

In order to have full control of the total iterations (total number of model runs) there will be more control on the total amount of run time.

This feature is requested by users calibrating the river bed roughness of the full Rijn in the Netherlands.

Be careful in adding all kind of magic parameters. What is the real issue they are running into casing the non convergence?
Note1: we have a restart functionality extremely useful for large expensive calibrations
Note2: restarting DUD from scratch (but starting at the latest best guess) after some outer iterations sometimes yields magic results (The affine approximation can become very poor after some updated in approximately the same search direction).