Set Up an Analysis Driver

In order to make a Dakota study and simulation model talk to each other, we must provide a mapping between the simulation model’s parameters and responses, and the Dakota study’s variables and responses, which may not be one-to-one.  To do this, we need to use the Analysis Driver wizard.

To create a new analysis driver, click on the New Analysis Driver link below the Analysis Drivers table on the settings editor view for your project:

DakotaGuiTutorialImage28.png

The Analysis Driver Wizard will appear.  Click past the first page, which only provides introductory information.  On the next page, select the simulation model and Dakota study that you want to have talk to each other:

DakotaGuiTutorialImage29.png

After making selections, click Next.

DakotaGuiTutorialImage30.png

The Analysis Driver Parameter Mapping page allows you to select Dakota variables that need to be mapped to the appropriate model parameters.  To access the available Dakota variables, click in an empty space in one of the rows under the Mapped Dakota Variable column.  You will see a dropdown appear that lists the available Dakota variables.

You are not required to map every model parameter and can leave some rows blank.  Suppose your Dakota study only defines three variables – “L”, “w”, and “t.”  In that case, your page may look something like this:

DakotaGuiTutorialImage31.png

When you’re ready to proceed, click Next.

DakotaGuiTutorialImage32.png

The Analysis Driver Response Mapping is an analogous page for mapping simulation model responses to Dakota responses.  Note that Dakota will not be able to run correctly if the expected number of model responses does not match the number of defined Dakota responses.

DakotaGuiTutorialImage33.png

The final page of the analysis driver wizard allows you to name your driver.  By default, the template for this driver (i.e. all the mapping instructions you’ve provided in this wizard) are stored in your Dakota project once you click Finish, but right now, this template is not usable by itself in a GUI-based workflow.  In order to use your driver in a workflow, you must export it as a Python script.  Click the checkbox for exporting as a Python script and click Finish.  You will be presented with a dialog that explains how to connect your Dakota study to the driver script you just generated:

DakotaGuiTutorialImage34.png

For instance, your Dakota’s interface block should look something like this:

DakotaGuiTutorialImage35.png

At this point, if you wish to add more sophisticated mapping logic to your driver script, you may open the Python script in Dakota GUI’s text editor by double-clicking on your Python driver script in the project navigator view.