Publication: Análisis de un modelo predictivo basado en Google Cloud y TensorFlow
Loading...
Files
Official URL
Full text at PDC
Publication Date
2017
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
Con este trabajo se propone un ecosistema de herramientas que permitan analizar y desarrollar modelos predictivos utilizando técnicas de deep learning, en particular redes neuronales. Con la ayuda de TensorFlow, una librería diseñada para resolver problemas de aprendizaje automático, se desarollará en local un benchmark que permita evaluar diferentes topologías de redes neuronales sobre un conjunto de datos.
Además, con el objetivo de poder analizar modelos en la nube utilizando grandes cantidades de datos, completaremos el ecosistema con dos herramientas de Google Cloud Platform: Datalab y ML Engine. La primera de ellas nos permitiría realizar una exploración y análisis inicial de datasets, incluyendo entrenamientos con redes neuronales para muestras de datos. Por otro lado, con ML Engine podremos realizar en la nube entrenamientos con datos a gran escala utilizando modelos Wide and Deep, los cuales combinan las ventajas de las regresiones lineales y no lineales.
A lo largo del trabajo se proporcionarán ejemplos y código para saber cómo trabajar con las distintas herramientas propuestas. En particular, se analizará un modelo predictivo con el objetivo de discernir transferencias fraudulentas en tarjetas de crédito.
This bachelor thesis is intended to introduce a set of tools in order to analyze and develop predictive models using deep learning techniques, in particular neural networks. Thanks to TensorFlow, a framework designed to solve machine learning problems, a local benchmark will be coded, so different neural networks can be tested on a dataset. What is more, in order to achieve cloud analysis on models using massive datasets, a couple of Google Cloud Platform services will be added up to the tools presented: Datalab and ML Engine. The first one will allow us firstly to explore and analyze datasets, including neural networks trainings for sampled data. On the other hand, ML Engine will let us train large-scale datasets using Wide and Deep models, which benefit from both linear and non linear regressions. Samples and code will be presented throughout the whole thesis, in order to know how to use the proposed tools. Particularly, we will analyse a predictive model that detects fraudulent transactions in credit cards.
This bachelor thesis is intended to introduce a set of tools in order to analyze and develop predictive models using deep learning techniques, in particular neural networks. Thanks to TensorFlow, a framework designed to solve machine learning problems, a local benchmark will be coded, so different neural networks can be tested on a dataset. What is more, in order to achieve cloud analysis on models using massive datasets, a couple of Google Cloud Platform services will be added up to the tools presented: Datalab and ML Engine. The first one will allow us firstly to explore and analyze datasets, including neural networks trainings for sampled data. On the other hand, ML Engine will let us train large-scale datasets using Wide and Deep models, which benefit from both linear and non linear regressions. Samples and code will be presented throughout the whole thesis, in order to know how to use the proposed tools. Particularly, we will analyse a predictive model that detects fraudulent transactions in credit cards.
Description
Trabajo de Fin de Grado en Ingeniería Informática y Matemáticas (Universidad Complutense, Facultad de Informática, curso 2016/2017)