Publication: Librería para la predicción usando el k-NN: paralelización y visualización de resultados
Loading...
Files
Official URL
Full text at PDC
Publication Date
2018
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
Desde hace unos años, cada vez se oye hablar más y más de Redes Neuronales, y toda clase de métodos complejos de clasificación y predicción dando la apariencia de que se restringen a estos exclusivamente. Sin embargo, existen otros muchos métodos de predicción y clasificación que sin ser tan complejos como los anteriores son igualmente válidos y puede dar resultados igual de buenos.
Este trabajo tiene por objetivo utilizar un método de los del segundo grupo. Concretamente utilizaremos el método de los k vecinos más cercanos, el cuál será explicado en detalle más adelante. El principal objetivo será arrojar luz sobre este método, dada su naturaleza de tipo “caja negra”, para poder entender mejor sus resultados. El trabajo tendrá dos partes:
• Desarrollar una serie de funciones para implementar el algoritmo y optimizar sus parámetros.
• Desarrollar una interfaz para visualizar sus resultados e interpretar los parámetros más fácilmente.
En esta memoria se describe todo el proceso que se ha llevado a cabo, las distintas iteraciones de las funciones, los paquetes utilizados, las conclusiones del desarrollo y una serie de mejoras que no se han planteado pero no han podido llevarse a cabo en el tiempo que dura el curso.
In the last years, we have heard more and more about Neural Nets, and all kinds of complex methods for classification and prediction. This gives the appearance that these are the only available options. Nevertheless, there exists a variety of other methods for prediction and classification without being of the complexity of the before mentioned and they are just as valid and give as good results. This work has the objective of utilizing a method within the second group. Our focus is on the k-nearest neighbors method. This will be explained in detail later. The main objective will be to shine a light on this method, given its “black-box” nature, in order to better understand the results. This work will have two parts: • Developing a series of functions to implement the algorithm and to optimize the parameters. • Developing a graphical user interface for viewing its results and easily interpreting the parameters. In this text we describe the whole process that was carried out, the packages utilized, the conclusions, and a series of improvements that could be done in the future.
In the last years, we have heard more and more about Neural Nets, and all kinds of complex methods for classification and prediction. This gives the appearance that these are the only available options. Nevertheless, there exists a variety of other methods for prediction and classification without being of the complexity of the before mentioned and they are just as valid and give as good results. This work has the objective of utilizing a method within the second group. Our focus is on the k-nearest neighbors method. This will be explained in detail later. The main objective will be to shine a light on this method, given its “black-box” nature, in order to better understand the results. This work will have two parts: • Developing a series of functions to implement the algorithm and to optimize the parameters. • Developing a graphical user interface for viewing its results and easily interpreting the parameters. In this text we describe the whole process that was carried out, the packages utilized, the conclusions, and a series of improvements that could be done in the future.
Description
Universidad Complutense, Facultad de Informática, curso 2017/2018