Automatic model search procedures aim at identifying the model that maximises a given fitness function, thereby treating model selection as an optimisation problem. However, it is unrealistic to believe that the fittest model represents the best solution to the search problem. In fact, even if it is possible to score all of the candidate models, it hardly happens that there exists an unequivocal answer to the question of which model best explains data. An automatic model search procedure for the identification of an optimal set of good models is proposed. In a technological approach to model selection the identified models can co-exist, whereas in a scientific modelling approach such models represent a starting point for further context-dependent analysis. An example of the application of the proposed procedure to real data is given.
Technological modelling for graphical models: an approach based on genetic algorithms / Roverato, Alberto; Paterlini, Sandra. - In: COMPUTATIONAL STATISTICS & DATA ANALYSIS. - ISSN 0167-9473. - STAMPA. - 47:2(2004), pp. 323-337. [10.1016/j.csda.2003.11.006]
Technological modelling for graphical models: an approach based on genetic algorithms
ROVERATO, Alberto;PATERLINI, Sandra
2004
Abstract
Automatic model search procedures aim at identifying the model that maximises a given fitness function, thereby treating model selection as an optimisation problem. However, it is unrealistic to believe that the fittest model represents the best solution to the search problem. In fact, even if it is possible to score all of the candidate models, it hardly happens that there exists an unequivocal answer to the question of which model best explains data. An automatic model search procedure for the identification of an optimal set of good models is proposed. In a technological approach to model selection the identified models can co-exist, whereas in a scientific modelling approach such models represent a starting point for further context-dependent analysis. An example of the application of the proposed procedure to real data is given.Pubblicazioni consigliate
I metadati presenti in IRIS UNIMORE sono rilasciati con licenza Creative Commons CC0 1.0 Universal, mentre i file delle pubblicazioni sono rilasciati con licenza Attribuzione 4.0 Internazionale (CC BY 4.0), salvo diversa indicazione.
In caso di violazione di copyright, contattare Supporto Iris