Implementación y prueba de una plataforma de Aprendizaje Automático a través de la librería Open Source MLflow

Abstract

Abstract: Machine learning is gaining strength every day in the field of data science, this allows reducing time and costs when developing projects, to achieve this There are multiple tools and that is why the proposal for this work was born, whose objective was; Implement and test a machine learning platform through the Open library Source MLflow. For this, the state of the art was carried out following the guide for Review Systematics of Literature (RSL) proposed by Kitchenham, selecting three works related projects whose results are positive when using the MLflow library. Likewise, it was used the machine learning life cycle as a basis for the methodology, but making emphasis on the main components that are Tracking, Projects and Models. This allowed perform the various functional tests of the tool from the installation of libraries until the execution of the local MLflow platform, where the results obtained in each training of the models and compare them, in order to take a decision of the most optimal model. To keep track of and track each change that was made to the tool is of great help, because it saved each of the executions carried out and adjustments that were made, with this we were able to make the different graphical and numerical comparisons between parameters and metrics used.