2.15.3. Configuring the MOP3 Node

To open the configuration dialog box, double-click the node or right-click and select Edit from the context menu.

Auto Button

The Auto button enables an automatic configuration of the mop competition and underlying models, depending on the given dataset. If some settings are changed on the Models or Competition tabs, the Auto button will be disabled automatically.

Overview Tab

The Overview tab provides information about the chosen database and the used models configured for the mop competition. Here you will find the number in use of:

Input parameters
Scalar responses
Signal responses
Successful designs
Incomplete designs
Models in competition

Auto Tab

This tab allows you to perform an automatic configuration of the settings in terms of models in competition, competition settings and FMU export. According to the selected importance of the desired Model quality and the spent Training time, a set of models in competition and the competition settings are automatically selected. If Enable FMU export is selected, only models which provide a FMU export are considered. Also, the FMU Export will be selected immediately.

Dimensions Tab

This tab allows you to select which parameters and responses will be used to build the MOP3. Select the parameters you want to use by clicking them and setting the Importance:

Mandatory: MOP3 forced to use the parameter
Selectable: MOP3 determines whether the parameter is relevant or not
Unimportant: MOP3 does not use the parameter

Select the relevant responses by selecting the corresponding check boxes.

Models Tab

This tab allows you to select and configure the model types you want to use in your metamodel competition.

The Available Models pane gives you a list of the known model types. The standard model types available with the MOP3 node are:

Kriging
Moving Least Squares
Linear Regression
Radial Basis Function
LSOPT FFNN Model
LSOPT Kriging Model
LSOPT RBF Model
GARS Model
SVR Model

To add a new model, click next to its name. A new instance of this model type is displayed in the Models in competition pane.

The model instances in the list will be used for your training of the MOP3. You may use more than one model instance of the same type. To keep track of them, give them a meaningful description by editing their names. To remove a model, click .

If you select a model instance the Model Configuration pane displays the model settings. You can change the settings to adapt the model to your needs.

Setting	Description
Kriging Models The Kriging model type is a regression model that interpolates the given input/output data by using Gaussian processes. You can configure the following settings.
Kernel	Select the kernel you want to train the Kriging model with. Gauss Anisotropic Gauss Isotropic Auto – The MOP3 automatically generates Kriging model instances with anisotropic and isotropic kernels
Value Transformation	Select if the input values are transformed during the training of the model. None – No transformation BoxCox – BoxCox transformation Auto – The MOP3 automatically generates Kriging model instances with none and BoxCox transformations
Use subspace filter	If activated, the MOP subspace filter is considered within the model assessment. If this is not the case, e.g. if the model has its own filter, then the model is evaluated in the full space of mandatory and selectable inputs.
Moving Least Squares Models The Moving Least Squares model type is a regression model that reconstructs the function of selected inputs/outputs by minimizing a weighted least-squares measure on polynomial functions. You can configure the following settings.
Kernel	Select the used kernel function for the MLS model. Gauss (smoothing) Regularized (interpolating) Auto – The MOP3 automatically generates MLS model instances with gauss and regularized kernels
Max Order	Specifies the maximal order/degree polynomial functions in the regression process may have. This value applies only if the Order setting is set to Auto.
Order	Sets the order of polynomial functions used in the model instance. Non-negative integer Auto – The MOP3 automatically generates Moving Least Squares model instances for all non-negative orders up to including Max. Order
Value Transformation	Sets if the input values are transformed during the training of the model. None – No transformation BoxCox – BoxCox transformation Auto – The MOP3 automatically generates Moving Least Squares model instances with none and BoxCox transformations
Use subspace fillter	If activated, the MOP subspace filter is considered within the model assessment. If this is not the case, e.g. if the model has its own filter, then the model is evaluated in the full space of mandatory and selectable inputs.
Linear Regression Models The Linear Regression model is trying to find a polynomial functional relation in the input/output data given to the MOP.
Mixed Term Interaction	Sets if the polynomial model also uses mixed terms (interaction between input variables in its definition). True – Consider the interaction of inputs during training False – Do not consider the interaction Auto – The MOP3 automatically generates Polynomial model instances for using the interaction and not using the interaction.
Max Order	Specifies the maximal order/degree polynomial functions in the regression process may have. This value applies only if the Order setting is set to Auto.
Order	Sets the order of polynomial functions used in the model instance. Non-negative integer Auto – The MOP3 automatically generates Polynomial model instances for all non-negative orders up to including Max. Order
Value Transformation	Sets if the input values are transformed during the training of the model. None – No transformation BoxCox – BoxCox transformation Auto – The MOP3 automatically generates Polynomial model instances with none and BoxCox transformations
Use subspace filter	If activated, the MOP subspace filter is considered within the model assessment. If this is not the case, e.g. if the model has its own filter, then the model is evaluated in the full space of mandatory and selectable inputs.
Radial Basis Function Models The Radial Basis Function Model is a regression model that uses a set of radial functions and a polynomial part to approximate the dependency between inputs and outputs. This model is currently under development.
Kernel	Select the kernel you want to train the RBF model with. Gauss Anisotropic Gauss Isotropic Auto – The MOP3 automatically generates RBF model instances with anisotropic and isotropic kernels
Value Transformation	Sets if the input values are transformed during the training of the model. None – No transformation BoxCox – BoxCox transformation Auto – The MOP3 automatically generates Polynomial model instances with none and BoxCox transformations
Number of alternate supports	Defines a threshold on the error during training that the Radial Base Functions models will allow. If the error of two model lies below this threshold, they are considered the same.

Each model is tagged with license information to show you what license is required to perform the MOP3 computation. You can configure the MOP3 without having this license, but you will need the license to execute the MOP3 node.

Competition Tab

This tab controls how the models selected for a competition are trained and compared against each other in order to bind the MOP.

Setting	Description
Quality measure	Select the quality measure used to define the optimality of a model in the competition. You have the following choices: CoP – Coefficient of Prognosis (default) CoD – Coefficient of Determination CoD adjusted – The adjusted version of the CoD
Testing type	Defines how the validation of the given samples is performed. Cross Validation – Separate the Set of input samples into k folds. The k is given by the Number of Folds setting Leave One Out – For n samples generate n training sets of n-1 samples validating with the one sample not in the training set Verification Data – Allows you to supply the verification set Custom – Undefined
Number of folds	When performing a cross validation training, you can specify through this setting the number of sets the input/output samples are separated in. Using five folds creates five training sets of 80% training samples and 20% validation samples.
Maximum number of responses in parallel
Error tolerance model	Defines the error difference threshold below which the quality measures of two models are considered equal. Difference is then defined by model complexity, meaning in a competition of two models with a quality measure difference below tolerance the lesser complex one wins.
Error tolerance parameter
Use uniform resampling	Turns on/off the uniform resampling during the computation for the Sobol-Indices for the Quality Measures.
Use incomplete designs	If selected, every available output data of the database is used for model training. If deselected, all incomplete designs are excluded completely for model training.
Use subspace filter	Turns on/off the filtering of subspaces of the model's domain that are deemed unimportant.
Use input correlation filter	Turn on/off the handling of highly correlated inputs.
Maximum input correlation

FMU Export Tab

This tab handles the settings and specifications of a possible FMU export of the resulting model after the MOP3 build is finished.

Report Tab

On this tab the results of the MOP3 for every output and every model in competition is shown.