To use the optimizer component you need to do the following steps:
-
In the algorithm tab you can choose an optimization algorithm that fits your problem. There are several algorithms from the Dakota package available.
There are properties for each algorithm. For editing, click on the 'Algorithm Properties' button. The appearing dialog shows the properties of the chosen algorithm. The properties differ from algorithm to algorithm. For more information about the properties see the documentation of the package.
If you have an operating system, on which the default Dakota does not work, you have the option to choose a custom Dakota binary by checking the box 'Use custom dakota binary'. You will be asked for the Dakota executable path at workflow start. This can be either a downloaded version from the dakota website or a self compiled binary with the source code from the Dakota website. For more information about compiling Dakota see: https://software.sandia.gov/trac/dakota/wiki/Developer
-
The next step is to define the inputs and outputs for the component. There are three types of data you can configure.
-
The objective function variables are the one to be optimized. For each variable you can specify if it should be minimized, be maximized or be searched for a specific value (solve for). If you have more than one objective function, you can define their weight in the optimization process. If there is only one objective function, the weight will be ignored. Note that some algorithms support single- and multi-objective optimization.
If you have defined some design variables, you also can choose if the objective function you create has gradients or not. If you select this, new inputs will appear in the connection editor, which are intended for the values from the gradients. Note that for every design variable you have, a new input for the objective function exists.
-
The constraint variables are used to bound particular variables to a region or value. If a solution is found but it causes a constraint variable to be out of bounds, the solution is not valid. Again, the constraints can have gradients.
-
The design variables are the values that are modified by the algorithm to find an optimal solution. For each variable you have to define a startvalue which will be the initial value for the optimization algorithm. You also have to define the lower and upper bound for each variable.
All data have to be either a float or a vector data type.
Note
Some Dakota algorithms do not support discrete optimization. For discrete design variables the following algorithms are available:
-
Dakota Coliny Evolutionary Algorithm
-
Dakota Multi Objective Genetic Algorithm
-
Dakota NOMAD
-
Dakota Single Objective Genetic Algorithm
All other Dakota methods will ignore discrete design variables during optimization.
There are three more tables for endpoints. They are just read-only and can not be modified. They are created automatically when configuring the Optimizer.
-
The 'Start value inputs' table shows which design variables need start values before running. Start values can be the starting value for this design variable, if the option 'Has start value' in the design variable dialog is not chosen.
Other possible start values are the lower and upper bounds of a design variable, if the option 'Has unified bounds' is not selected.
-
In the 'Optimal design variables outputs' table, the outputs for the values of the optimized point are shown.
-
The 'Gradients' table shows which objective functions should have gradient inputs as well. This is chosen in the dialog for the objective functions ('Has gradient').
-
After these steps the optimizer component is ready to start. In a running workflow, you are able to see the output from the Dakota optimizer in the Workflow console.
By double clicking the Optimizer component in the runtime view of the workflow, you will get the values the optimizer produces and the possibility to export these values to an Excel file. You are also able to plot a graph with the given results in the diagram tab.
For help concerning nested and fault-tolerant loop settings, see the section 'Usage/Workflows' in the user guide.