Configuring the Design Variables

The list of parameters includes all input and in/out parameters from the Optimization adapter and parameters of the same modes from the subflow adapters, if they are not mapped to any parameters of the Optimization adapter.

1. Determine which parameters you want to use as design variables by selecting the corresponding check boxes in the first column. Alternatively, you can click Select all to add all the parameters (including array elements) as design variables. To clear all the parameters, click Clear selection.

   If no parameters are selected, you will be prompted to add all parameters as design variables.

   Selecting a subflow parameter as a design variable will, at a later time, create a corresponding parameter in the Optimization adapter. Only real, integer, and string parameters can be design variables.

2. Specify the Lower Bound for the design variable in the corresponding column. If you do not specify lower bound values, Optimization Process Composer uses the default value $1\times {10}^{15}$. This results in a large range for the design variables and may reduce the chances of finding a good design. It is recommended that you specify lower bounds for all design variables. This setting is required if you are using the automatic scaling adapter preference.
3. Change the initial value in the Value column.
   Note: This column is not available for the Multi-Island Genetic Algorithm because the initial design point is not used by this algorithm.

   Warning: The initial values of the design variables are overridden at execution time if the Optimization adapter is a child adapter of another process adapter (for example, a DOE adapter). Therefore, to change the initial values of design variables, you must change the corresponding parameter values in the parent (DOE) adapter.

   Changing initial values of design variables changes their values in the main window.

4. Specify the Upper Bound for the design variables in the corresponding column. If you do not specify upper bound values, Optimization Process Composer uses the default value $1\times {10}^{15}$. This results in a large range for the design variables and may reduce the chances of finding a good design. It is recommended that you specify upper bounds for all variables. This setting is required if you are using the automatic scaling adapter preference.
5. Specify the Step Size for the design variable in the corresponding column. This value is the spacing between the potential sample points within the given bounds. The step size for real parameters must be any positive real value; the value for integer parameters must be any positive integer value. If you do not specify the Step Size for real parameters, the step size is the value of the tuning parameter Significant Step Size times the span of the bounds; if you do not specify the Step Size for integer parameters, the step size is 1.
6. Type a list of allowed values in the Allowed Values column, separated by a semicolon. If you enter allowed values, the lower and upper bounds are erased and vice versa. Entering a list of allowed values creates a discrete variable, and the Optimization adapter uses only these values during optimization. The optimization algorithm will view this variable as an integer parameter with the range of values $0,1,2,\mathrm{...},N-1$, where $N$ is the length of the allowed values list.
7. Set the Scale Factor for the variable in the corresponding column. Scale factors are used to bring variable values to the same order of magnitude to improve the efficiency of the optimizers.
   Note: This column does not appear if you use the automatic scaling adapter preference.

8. (Multi-Island Genetic Algorithm technique only) Set the value in the Gene Size column. This value controls the number of bits $N$ in all genes used for encoding the value of each variable. Every bit of the gene can change its value between 0 and 1. The total number of possible combinations in every gene is then $2N$. This number of combinations determines the minimum change in the value of any design variable during all genetic operations—take the allowed range of values for a design variable and divide it by the total number of combinations. To increase the minimum change in design variable values (that is, to decrease the number of possible bit combinations when the allowed range of design variable values is small), decrease the gene size.