The Bayesian Optimization Report

The Bayesian Optimization report contains a Model Summary tab, a Batch Selection tab, and a tab for each response variable.

Figure 18.5 The Bayesian Optimization Report Tabs 

Model Summary Tab

The Model Summary tab contains the Bayesian Optimization Model Summary report. This report provides a summary of the information in the individual response reports that are generated for each response that is specified in the launch window. The report includes model diagnostics, actual by predicted plots, and prediction profilers for all of the responses. In the profilers section, there is a button that enables you to show or hide plots of the response prediction standard deviations. By default, plots for the standard deviations are not shown. For more information about the prediction profiler, see “Profiler” in Profilers.

The Diagnostics Summary table contains the Leave-One-Out RSquare and Measurement Error values for each individual Gaussian Process fit. Models with an RSquare value below the threshold value that is specified in the launch window are highlighted in red. The Measurement Error is calculated as the product of the nugget parameter, τ, and the Gaussian Process variance, σ2.

Batch Selection Tab

The Batch Selection tab contains a Bayesian Optimization Batch Customizer section, Candidate Set section, and a Current Batch section. The Bayesian Optimization Batch Customizer section is closed by default. This section enables you to explore different ways to generate new factor settings and contains the current candidate set. The Current Batch section lists the factor combinations that are selected to be included in the current batch and options regarding where to save them. See Batch Selection Tab.

Response Tabs

There is a tab for each response that is specified in the launch window. Each response tab contains the results from a Gaussian Process model that is fit to the response. The parameter estimates for the factors are provided, as well as estimates for the intercept, β, nugget parameter, τ, and the Gaussian process variance, σ2. Leave-One-Out RSquare and Measurement Error metrics are also provided. The Measurement Error is calculated as the product of the nugget parameter, τ, and the Gaussian Process variance, σ2. Each report also contains an actual by predicted plot and a prediction profiler. For more information about the prediction profiler, see “Profiler” in Profilers. For more information about Gaussian Process models, see “Gaussian Process”.