1Blurb:: 2Compute surrogate quality metrics 3 4Description:: 5Diagnostic metrics assess the goodness of fit of a global surrogate to 6its training data. 7 8The default diagnostics are: 9\li \c root_mean_squared 10\li \c mean_abs 11\li \c rsquared 12 13Additional available diagnostics include 14\li \c sum_squared 15\li \c mean_squared 16\li \c sum_abs 17\li \c max_abs 18 19The keywords \c press and \c cross_validation further specify 20leave-one-out or k-fold cross validation, respectively, for all of the 21active metrics from above. 22 23<b>Usage Tips</b> 24When specified, the \c metrics keyword must be followed by a list of 25quoted strings, each of which activates a metric. 26 27Topics:: surrogate_models 28 29Examples:: 30This example input fragment constructs a quadratic polynomial 31surrogate and computes four metrics on the fit, both with and without 325-fold cross validation. (Also see 33dakota/share/dakota/test/dakota_surrogate_import.in for additional 34examples.) 35 36\verbatim 37model 38 surrogate global 39 polynomial quadratic 40 metrics = "root_mean_squared" "sum_abs" "mean_abs" "max_abs" 41 cross_validation folds = 5 42\endverbatim 43 44Theory:: 45 46Most of these diagnostics refer to some operation on the 47residuals (the difference between the surrogate model and the truth 48model at the data points upon which the surrogate is built). 49 50For example, \c sum_squared refers to the sum of the squared 51residuals, and \c mean_abs refers to the mean of the absolute value of 52the residuals. \c rsquared refers to the R-squared value typically 53used in regression analysis (the proportion of the variability in the 54response that can be accounted for by the surrogate model). Care 55should be taken when interpreting metrics, for example, errors may be 56near zero for interpolatory models or rsquared may not be applicable 57for non-polynomial models. 58 59Faq:: 60See_Also:: 61