| Model-Based Calibration Toolbox™ | |
| On this page… |
|---|
The option Build Models (
) is
in the toolbar and the Model menu for global
models, and in the Model menu for local
models. You can use this to quickly build a selection of models as
child nodes of the current global model node, or in the case of local
models, to build a selection of child nodes for each response feature
node. You can use predefined templates, model types in the current
project, create templates within the Build Models dialog box, or create
templates from a model with existing child nodes by selecting Model
> Make Template.
Once you have used the Build Models dialog box to create a number of child nodes, the best model of this selection will be automatically selected (it will have a blue icon), based on the selection criteria you choose (such as PRESS RMSE, RMSE, Box-Cox, or Parameters). For local nodes, the best child node of each response feature is chosen.
See the Quick Start tutorial section Creating Multiple Models to Compare in the Getting Started documentation for a step by step example of using the Build Models dialog box to quickly create many models to search for the best fit.
Note Once you start building models from any template, you can always click Stop to abort model building if the process is taking too long. |
To create a template containing the models you want to build,
Click New then click OK.
Click Add in the following Multi-Model Settings dialog box to reach the Model Setup dialog box, where you can select any model type available for the number of inputs. Click OK to add the model and return to the Multi-Model Settings dialog box.
Click Add again to repeat the process to add as many different models as you like. You can click Edit Model to change the settings for any models in the list.
When you are satisfied, click OK in the Multi-Model Settings dialog box, and your chosen selection of model types are built as a selection of child nodes of the currently selected model node.
The Model Selection dialog box appears, where you can select the criterion for automatically choosing the best child node. Use the drop-down menu to select from the available criteria (such as from PRESS RMSE, RMSE, Box-Cox, Observations or Parameters). You can select additional criteria to appear here using the Summary Statistics options, from the Model menu. See Model Menu (Global Level) and Summary Statistics. Click OK to accept the chosen criterion.
You can save templates of models you have already built by using the Model > Make Template menu item from any global modeling node with child nodes. User-defined templates can then be found via the Build Models dialog box. You can use the Browse button to find stored templates that are not in the default directory. Select your template and click OK. The models are built and the Model Selection dialog box appears, where you can select the criterion for automatically choosing the best child node.
You can use the polynomial template to build a variety of polynomials of different orders.
Select Polynomials and click OK. The Model Building Options dialog box opens where you can specify the model settings.
Choose the minimum and maximum order of the polynomials you want to build, and whether to use Stepwise settings. For example, if you choose 1 and 5 as the minimum and maximum polynomial order, 5 child node models are built (linear, quadratic, cubic, and so on). If you choose a Stepwise setting (e.g. Minimize PRESS) it is applied to all child models.
Click Build and the models are built. The Model Selection dialog box appears, where you can select the criterion for automatically choosing the best child node.
You can use the RBF template to build a variety of radial basis function models with varying maximum numbers of centers and/or different kernels.
Select RBF and click OK. The Model Building Options dialog box opens where you can specify the model settings.
Enter a vector in the edit box to specify the maximum numbers of centers for each child model. This can be a MATLAB® expression including the number of observations, e.g. 10:10:nObs/2.
If the current model node is an RBF, the same model settings are used by default. Click Model Settings to open the Radial Basis Function Model Settings dialog box, where you can view and change all the model parameters such as kernel and widths. See Types of Radial Basis Functions.
Select the check box Build all kernels to create child models with the specified range of centers for each kernel type.
Click Build and the models are built. The Model Selection dialog box appears, where you can select the criterion for automatically choosing the best child node.
You can use the Hybrid RBF template to build a variety of hybrid radial basis function models with varying maximum numbers of centers and/or different kernels.
Select Hybrid RBF and click OK. The Model Building Options dialog box opens where you can specify the model settings.
Enter a vector in the edit box to specify the maximum numbers of centers for each child model. This can be a MATLAB expression including the number of observations, e.g. 10:10:nObs/2.
If the current model node is a hybrid RBF, the same model settings are used by default. Click Model Settings to open the Hybrid RBF Model Settings dialog box, where you can view and change all the model parameters such as kernel and widths (and the order of the polynomial part of the model on the Linear Part tab). See Hybrid Radial Basis Functions.
Select the check box Build all kernels to create child models with the specified range of centers for each kernel type.
Click Build and the models are built. The Model Selection dialog box appears, where you can select the criterion for automatically choosing the best child node.
You can use the Free Knot Spline template to build a variety of free knot spline models of different numbers of knots. Only available for models with a single input factor.
Select Free Knot Spline and click OK. The Model Building Options dialog box opens where you can specify the model settings. If the current model node is a free knot spline, the same model settings are used by default.
Choose the initial and final number of knots. For example if you specify the initial and final numbers of knots as 1 and 5, five child nodes are built, one with one knot, one with two and so on.
Click Build and the models are built. The Model Selection dialog box appears, where you can select the criterion for automatically choosing the best child node.
You can use the Model Browser template to build a copy of an existing set of child model types in the current project. This avoids the need to create a model template first.
Select Model Browser and click OK.
The Model Tree dialog box opens. Select a model from the model tree which has the child node model types you want to build. You can also use this template to select a local multiple model node to copy. Click OK to return to the Build Models dialog box.
Click Build and the models are built. The Model Selection dialog box appears, where you can select the criterion for automatically choosing the best child node.
If you have Parallel Computing Toolbox™ available, you can distribute model building tasks to a cluster of computers. The models are then built in parallel. This option can significantly reduce the computation time for building large numbers of models, as long as each model fit is taking a lot longer than the time it takes to send the problem to another computer.
Use the function mbcmatlabpool to configure the toolbox to fit models using the parallel language worker pool.
Start the parallel language worker pool (using matlabpool),
Enter mbcmatlabpool open
The following model fitting tasks are run in parallel:
Build models from model templates
Update response or response feature models after changing data
Note No progress bars display when fitting multiple models in parallel. |
Enter mbcmatlabpool close to turn off the parallel execution and enable display of progress during the build for modelling tasks. The status of mbcmatlabpool persists between MATLAB sessions.
You can query the status of mbcmatlabpool by entering status = mbcmatlabpool.
See also the functions matlabpool, parfor.
Some other model-building tasks (such as building alternative models for local multiple models) automatically use the parallel language worker pool, if it is open.
Note Opening matlabpool may prevent other jobs (e.g., distributed optimizations in CAGE) from being processed. |
| New Response Models and Datum Models | Selecting Models | |
| © 1984-2008- The MathWorks, Inc. - Site Help - Patents - Trademarks - Privacy Policy - Preventing Piracy - RSS |