Modelos multivariados de covariância linear generalizada em Python : a biblioteca MCGLM
Resumo
Resumo: O advento da computação é fundamental ao desenvolvimento da Estatística; a Computação viabilizou modelos estatísticos, impulsionou a pesquisa e permitiu a distribuição de bibliotecas em larga-escala. Não obstante, a ciência Estatística transmutou a Computação, transformando-na em um elemento intransponível da sociedade moderna. Esta dissertação é mais um evento dessa amálgama próspera. Modelos Multivariados de Covariância Linear Generalizada é uma família de modelos estatísticos, postulada pelos pesquisadores Wagner Bonat e Bent Jørgensen, que universaliza o Modelo Linear Generalizado (GLM) para a análise multivariada de dados dependentes. Esta dissertação se propôs a implementar e publicar esse modelo estatístico para a linguagem Python em código aberto, como uma biblioteca. Como resultado, o código fonte está disponivel em: https://github.com/jeancmaia/mcglm e a biblioteca pode ser visualizada no PyPI, o principal repositório de software da linguagem Python, através do link: https://pypi.org/project/mcglm/. A biblioteca Python mcglm foi implementada nas normas da programação orientada a objetos, amparada nos princípios do acronimo SOLID, de Robert C. Martin. O desenvolvimento do código foi conduzido por testes unitários, perfazendo em uma cobertura de mais de noventa e um por cento. Visando uma fácil adesão à bibilioteca, as interfaces de treinamento e análise replicam o padrão da biblioteca statsmodels, a principal ferramenta para análises estatísticas na linguagem Python. Como objeto desta dissertação, os usuários da linguagem Python tornam-se aptos a acessar a um modelo estatístico inovador, oportunamente testado e plenamente extensível. A biblioteca mcglm amplia a lista de algoritmos estatísticos existentes na linguagem Python e ajuda a promover o aprendizado estatístico para os usuários desta linguagem, uma das mais expressivas da década. Abstract: The advent of computing is pivotal in the history of Statistics; Computing enabled statistical models, boosted research and enabled large-scale distribution of libraries. Nevertheless, Statistical science has transmuted Computing, transforming it into an insurmountable element of modern society. This dissertation is one more event in this thriving amalgamation. Multivariate Covariance Generalized Linear Models is a family of statistical models postulated in 2015 by researchers Wagner Bonat and Bent Jørgensen. This family of models proposes an extension to the Generalized Linear Model (GLM) for multiple dependent data. This dissertation proposed to implement and publish this statistical model for the Python language in an open way, as a library. As a result, the source code is available at: https://github.com/jeancmaia/mcglm and the library can be viewed on PyPI, the main Python language software repository, via the link: https://pypi.org /project/mcglm/. The Python mcglm library was implemented in the norms of Object Oriented programming, supported by the principles of the acronym SOLID, by Robert C. Martin. Code development was driven by unit testing, yielding over ninety-two percent coverage. Aiming at an easy accession to the library, the training and analysis interfaces replicate the standard of the statsmodels library, the main tool for statistical analysis in the Python language. As the object of this dissertation, Python language users become able to access an innovative statistical model, timely tested and fully extensible. The mcglm library expands the list of statistical algorithms existing in the Python language and helps to promote statistical learning for users of this language, one of the most expressive of the decade.
Collections
- Dissertações [261]