Evaluating the performance of correlated methods in molecular property calculations: Pattern recognition and clustering in spaces of theoretical descriptions
George B. Maroulis · International Journal of Quantum Chemistry · 1995
Abstract A rigorous approach to the evaluation of the performance of correlated methods in molecular property calculations is proposed. Theoretical descriptions of molecules are identified as collections of molecular property values. Distance functions are then defined in the space of theoretical descriptions and the metric properties are used to define proximity and similarity between theoretical descriptions. Graph theoretic arguments and pattern recognition techniques are used to study ordering, classification, self‐organization, and clustering in the space of theoretical descriptions. © 1995 John Wiley & Sons, Inc.