Metodología de integración de modelos basados en datos en entornos de producción

Abstract

At present, there are numerous methodologies that describe what to do in the different stages of data modelling projects, but these do not describe how to do it or what architectures or technologies to use for. In practice, a multitude of heterogeneous technologies and tools coexist and are used ad hoc according to the needs of each model. In most cases, the processes of training, validation, deployment, monitoring and re-training of the models are carried out without a common technology framework that serves as a reference for all projects belonging to the same aim, which hinders the management of the models and their life cycle. The present work presents a proposal of methodology and architecture based on virtualization technologies, that serves as a guide for the correct execution of data modelling projects and that allows to manage the phases of data analysis, model development, production implementation, monitoring and testing of new configurations.