Score
The quality of each grid point is evaluated based on a score function, which plays the role of the objective function to be minimized in the parameter-optimization problem.
The default score function is a weighted root mean square difference (RMSD) between the estimated property values and their corresponding reference values,
where
the index \(i\) is the linear index of the grid point under evaluation;
the index \(p\) runs over the composite properties;
\(N_P\) is the number of composite properties;
\(\mathbf{w} = (w_1,\cdots,w_{N_P})\) are the weights of the properties (given in the input file);
\(\mathbf{r} = (r_1,\cdots,r_{N_P})\) are the reference values of the properties (given in the input file);
\(\mathbf{a}^{{(k)}} = (a_{pi}^{(k)})_{p}\) is the vector of surrogate-model-estimated values of the properties for the grid point \(i\).
For flexibility, one can define customized score functions by means of the customization API.