gemseo.disciplines.scenario_adapters package#

Scenario adapters.

A scenario adapter is an Discipline wrapping a Scenario. A call to Discipline.execute() triggers calls to Scenario.execute().

For instance, let us consider an MDOScenario defining a gradient-based constrained minimization of a cost function over a DesignSpace from several Discipline instances. If this optimization problem is not convex, it is advisable to set up a multi-start strategy to repeat this minimization from different starting points. in order to find a good local minimum. In this case, an MDOScenarioAdapter takes a design value as input, use it as initial design value of the minimization algorithm and outputs some variables of interest such as the objective and constraints at the optimum. Then, this MDOScenarioAdapter can be used as any Discipline in a DOEScenario defining a sampling-based version of the previous problem. In other words, this DOEScenario repeats the gradient-based optimization from several starting points and returns the best local minimum.

The scenario adapters can also be useful for bi-level optimization. Let us consider an optimization problem with two design variables, namely \(x_1\) and \(x_2\). The wrapped MDOScenario solves the optimization problem with respect to the design variables \(x_1\) and another MDOScenario considers this MDOScenarioAdapter to solve the optimization problem with respect to \(x_2\). It is particularly relevant when the design variables have different natures, e.g. \(x_1\) is discrete and \(x_2\) is continuous, and that dedicated algorithms exist.

Submodules#

On this page

This Page