Publication: Desarrollo y selección de características basadas en distancias entre grafos para problemas de predicción en planificación automática
Loading...
Identifiers
Publication date
2011-09
Defense date
2011-09-30
Authors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
El objetivo del presente proyecto es el desarrollo y la selección de características basadas en la distancia entre grafos para problemas de predicción en planificación automática. En la planificación automática existen dos componentes: los dominios y los problemas. De estos elementos se puede generar diversa información, como el número de objetos, número de predicados, la longitud de los planes que los resuelven o el número de nodos generados en el proceso de cómputo del plan. Además de esta información se va a utilizadar una serie de grafos que permiten comparar distintos problemas. Estos grafos se basan en la representación SAS+, que consiste en una lógica proposicional que posee variables de estado multivaluadas con el fin de expresar una estructura causal subyacente. Con todos los datos que se han obtenido, se ha creado una representación común independiente del dominio de planificación. Después se van a realizar una serie de comparaciones entre el problema a resolver y la base de hechos (problemas con solución), creando para ello unas medidas de distancia para los grafos. Estas comparaciones han sido almacenadas para su posterior recuperación. Todo el conocimiento generado se puede utilizar para generar modelos de predicción o para el razonamiento basado en casos. Y con estos modelos se ha comprobado si las medidas desarrolladas con los grafos son acertadas. Además se han realizado procesos de selección de características de los datos de los problemas, para concluir qué variables son las más relevantes para los problemas de planificación automática. Las principales conclusiones obtenidas de este proyecto son la aportación de las características más significativas, además de comprobar que la medida de distancia de los grafos es acertada para la creación de los modelos. ______________________________________________________________________________________________________________________________________
The objective of this project is to explain the development and selection of features based on the distance between graphs for prediction problems in automated planning. In automatic planning, there are two fundamental components, the domains and problems. Of these elements can generate various information such as the number of objects, number of predicates, the length of the plans that meet or the number of nodes generated in the process of computing the plan. Addition to this information, we used a series of graphs. These graphs are based on the representation SAS+, which is a propositional logic that has multi-valued state variables in order to express an underlying causal structure. With all the gathered data, a domain-independent representation has been developed. Then they will make a series of comparisons between two di fferent problems, thus creating some distance measures for graphs. These comparisons have been stored for later retrieval. All the knowledge generated can be used to generate predictive models or case-based reasoning. And with these models, wether the measures developed in the graphs are correct has been checked. In addition, feature selection processes have been developed, to conclude which variables are relevant to the problems of automatic planning. The most important conclusions from this project are providing the most signi cant features and to verify that the graph based distance measure are accurate for creating prediction models.
The objective of this project is to explain the development and selection of features based on the distance between graphs for prediction problems in automated planning. In automatic planning, there are two fundamental components, the domains and problems. Of these elements can generate various information such as the number of objects, number of predicates, the length of the plans that meet or the number of nodes generated in the process of computing the plan. Addition to this information, we used a series of graphs. These graphs are based on the representation SAS+, which is a propositional logic that has multi-valued state variables in order to express an underlying causal structure. With all the gathered data, a domain-independent representation has been developed. Then they will make a series of comparisons between two di fferent problems, thus creating some distance measures for graphs. These comparisons have been stored for later retrieval. All the knowledge generated can be used to generate predictive models or case-based reasoning. And with these models, wether the measures developed in the graphs are correct has been checked. In addition, feature selection processes have been developed, to conclude which variables are relevant to the problems of automatic planning. The most important conclusions from this project are providing the most signi cant features and to verify that the graph based distance measure are accurate for creating prediction models.
Description
Keywords
Teoría de grafos, Control automático, SAS (Programa de aplicación)