11.4.4. Prediction of the variance

Once the estimate is completed, it is possible to compute the central value for a new set of input values (i.e., for a new design-of-experiments) using the newly estimated parameter values. Although this applies to every method in the calibration module, the Linear Bayesian procedure has the advantage of providing the covariance matrix of the parameters. Under some assumptions on the input distribution, it is possible to obtain a variance for each new predicted value, reflecting only the uncertainty due to the parameters. For more details on the estimate, see [Bla17]. This can be done by calling the computePredictionVariance method:

void computePredictionVariance(URANIE::DataServer::TDataServer *tdsPred, string outname);

This method takes two arguments, which are:

1. tdsPred: a TDataServer containing the new locations to be estimated, in which all regressors must be available in order to be able to compute the covariance matrix;

2. outname: the name of the attribute to be created, which will be filled with the diagonal elements (the variances) of the \( \Sigma^{pred}_{\theta} \) matrix.