Submitted by Nathália Faria da Silva (nathaliafsilva.ufv@gmail.com) on 2017-07-13T11:23:31Z No. of bitstreams: 1 texto completo.pdf: 1310330 bytes, checksum: 0b1fb40f1985db830fa723ed7b82aec9 (MD5) / Made available in DSpace on 2017-07-13T11:23:31Z (GMT). No. of bitstreams: 1 texto completo.pdf: 1310330 bytes, checksum: 0b1fb40f1985db830fa723ed7b82aec9 (MD5) Previous issue date: 2001-11-13 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / The objective of the first study was to obtain genetic parameters for sequential weights of beef cattle using RRM on data sets with missing and no missing traits, and to compare these estimates with those obtained by MTM. Growth curves of Nellore cattle were analyzed using body weights measured at ages ranging from 1 day (birth weight) to 733 days. Two data samples were created: one with 71,867 records from herds with missing traits and the other with 74,601 records from herds with no missing traits. Records preadjusted to a fixed age were analyzed by a multiple trait model (MTM), which included the effects of contemporary group, age of dam class, additive direct, additive maternal, and maternal permanent environment. Analyses were by restricted maximum likelihood (REML) with 5 traits at a time. The random regression model (RRM) included the effects of age of animal, contemporary group, age of dam class, additive direct, additive maternal, permanent environment, and maternal permanent environment. Legendre cubic polynomials were used to describe the random effects. Estimates of covariances by MTM were similar for both data sets, although those from the missing data set showed more variability from age to age. The estimates from RRM were similar to those from MTM only for the complete -trait case and showed large artifacts for the case of missing traits. Estimates of additive direct-maternal correlations under RRM for some ages approached -1.0, and most likely contained artifacts. If many traits are missing, the best approach to obtaining parameters for RRM would be conversion from smoothed MTM estimates. The purpose of the second study was estimation of parameters of models and data sets as in the first study by a Bayesian methodology – Gibbs sampling, and to make comparisons with their estimates by REML. Analyses were by a Bayesian method for all 9 traits. MTM estimated covariance components and genetic parameters for birth weight and sequential weights and RRM for all ages. Estimates of additive direct variance from herds with missing traits increased from birth weight through weight at 551 to 651 days with MTM. However, this component also increased for the sample with no missing traits after this age. Additive direct and residual estimated variance with RRM increased over all ages for both samples. For MTM, additive direct and maternal heritabilities were greater from the sample with herds with missing traits than those values from herds with no missing traits. The estimates from RRM were slightly lower than those from MTM for the sample with no missing traits; however, additive maternal heritabilities from MTM were greater than those using RRM. The estimated additive direct genetic correlations for each pair of traits were slightly higher for the first age (birth weight) using MTM than RRM. The range of additive maternal genetic correlations was lower than that for additive direct genetic correlations with MTM and RRM. Due to the fact that covariance components based on RRM were inflated for herds with missing traits, MTM should be used and converted to covariance functions. As well, for analyses with standard models where inferences on shapes of parameters are not important, analyses by REML may be more robust. The first goal of the third study was to implement the genetic evaluation of weights for a large population of beef cattle using the random regression model. The second goal was to compare these evaluations with those obtained from a multitrait evaluation. Expected progeny differences (EPD) were computed by two methods: a finite method using sparse factorization (SF) and interating (IT) by preconditioned conjugate gradient (PCG). The correlations between EPDs from MTM and RRM by SF and IT were ≤ .43 until the random regressions were orthogonalized. After orthogonalization high computing requirements of RRM were reduced by removing regressions corresponding to very low eigenvalues and by replacing the random error effects with weights. Correlations between EPDs from MTM and RRM for the additive direct effect were .87, .89, .89, .87, and .86 for W1 (weight at 60 days), W2 (weight at 252 days), W3 (weight at 243 days), W5 (weight at 426 days), and W7 (weight at 601 days), respectively. The corresponding correlations for the additive maternal effect were .85, .86, .88, .85 and .84, respectively. These low correlations were mostly due to differences in variances between the models and, to a lesser degree, due to better accounting for environmental effects and more data by RRM. The RRM applied to beef weights may be poorly conditioned numerically. / O objetivo do primeiro estudo foi estimar parâmetros para pesos seqüenciais de gado de corte, por meio de modelos de regressão aleatória (RRM), em características com informações perdidas e completas. Analisaram-se curvas de crescimento de gado Nelore mediante o uso de pesos corporais coletados, do nascer aos 733 dias de idade. Duas amostras foram geradas; a primeira era constituída de 71.867 medidas provenientes de rebanhos com informações perdidas, e a segunda, de 74.601 medidas oriundas de rebanhos com informações completas. Os pesos pré-ajustados a idades fixas foram analisados por meio de um modelo de características múltiplas (MTM), cinco características por vez, no qual foram incluídos efeitos de grupo contemporâneo, classe de idade da vaca, aditivo direto, aditivo materno e ambiente materno permanente. No modelo de regressão aleatória (RRM) foram incluídos efeitos de idade do animal, grupo contemporâneo, classe de idade da vaca, aditivo direto, ambiente permanente, aditivo materno e ambiente materno permanente. Polinômios cúbicos de Legendre foram utilizados na descrição dos efeitos aleatórios. Estimativas de covariâncias por meio de MTM foram similares em ambas as amostras, apesar de as obtidas da amostra com informações perdidas terem apresentado maior variabilidade entre as idades. As estimativas obtidas pelo RRM foram similares às obtidas pelo MTM somente para o caso de características completas e mostraram grande variabilidade para o caso de características com informações perdidas. Estimativas de correlações entre os efeitos aditivos direto e materno, por meio de RRM, foram iguais a -1.0, em algumas idades. Se várias informações forem perdidas, a melhor aproximação para obter parâmetros por meio de RRM seria a conversão das estimativas obtidas por meio de MTM. O segundo estudo objetivou estimar parâmetros por meio de modelos e características com informações perdidas e completas, à semelhança do primeiro estudo, mediante metodologia Bayesiana – Gibbs sampling, e efetivar comparações com as estimativas obtidas por meio da metodologia REML. As análises por meio do MTM foram para nove características. Estimaram-se componentes de covariâncias e parâmetros genéticos para específicos pontos seqüenciais, por meio do MTM; entretanto, por meio do RRM, tais estimativas foram obtidas para todas as idades. Estimativas de variâncias aditivas diretas para a amostra com informações perdidas aumentaram, do nascer à idade de 551 a 651 dias, pelo MTM, e em todas as idades, na amostra com informações completas. Estimativas de variâncias aditiva direta e residual, mediante RRM, aumentaram ao longo de todas as idades, em ambas as amostras. Pelo MTM, heritabilidades aditivas direta e materna foram maiores na amostra de rebanhos com informações perdidas do que na de rebanhos com informações completas. As estimativas obtidas pelo RRM foram ligeiramente menores do que aquelas obtidas pelo MTM na amostra com informações completas. Heritabilidades aditivas maternas pelo MTM foram maiores do que aquelas obtidas pelo RRM. As estimativas de correlações genéticas aditivas diretas foram levemente maiores para peso ao nascer, quando se utilizou MTM do que quando se empregou RRM. A amplitude das correlações genéticas aditivas maternas foi menor do que a do efeito genético aditivo direto, pelo MTM e pelo RRM. Tendo em vista que os componentes de covariância baseados em RRM são influenciados por informações perdidas, recomendam-se o MTM e a conversão destes componentes em funções de covariância. Além disso, nas análises com modelos-padrão em que inferências dos parâmetros não são importantes, o REML deve ser escolhido. Um terceiro trabalho objetivou a implementação de avaliação genética em bovinos de corte, utilizando modelo de regressão aleatória. Além disso, as avaliações foram comparadas com aquelas estimadas por meio de um modelo de características múltiplas. Dois métodos foram considerados nas análises: um método finito, FSPAKF90 (Factorization sparse matrix package), e o de iteração nos dados, PCG ( Preconditioned conjugate gradient). As correlações entre as diferenças esperadas nas progênies (DEP), estimadas pelo MTM e pelo RRM, foram muito baixas antes de se terem as regressões aleatórias ortogonais. Grande demanda computacional dos RRM foi reduzida pela remoção das regressões correspondentes a pequenas variâncias e também pela substituição dos efeitos aleatórios do erro por específica ponderação. Correlações entre DEPs, estimadas pelo MTM e pelo RRM para efeito aditivo direto, foram .87, .89, .89, .87 e .86 para W1 (peso aos 60 dias), W2 (peso aos 152 dias), W3 (peso aos 243 dias), W5 (peso aos 426 dias) e W7 (peso aos 601 dias), respectivamente. As correlações correspondentes, para efeito aditivo materno, foram .85, .86, .88, .85 e .84, respectivamente. Estimativas obtidas pelos RRM em informações ponderais de gado de corte podem não ser adequadas, em virtude das propriedades numéricas desses modelos. Em geral, baixas correlações são devidas a diferenças em variâncias entre modelos, número insuficiente de graus de liberdade para estimar os efeitos de ambiente e informações perdidas nos RRM.

No description available.

The fully Bayesian estimation via the use of Markov chain Monte Carlo (MCMC) techniques has become popular for estimating item response theory (IRT) models. The current development of MCMC includes two major algorithms: Gibbs sampling and the No-U-Turn sampler (NUTS). While the former has been used with fitting various IRT models, the latter is relatively new, calling for the research to compare it with other algorithms. The purpose of the present study is to evaluate the performances of these two emerging MCMC algorithms in estimating two two-parameter logistic (2PL) IRT models, namely, the 2PL unidimensional model and the 2PL multi-unidimensional model under various test situations. Through investigating the accuracy and bias in estimating the model parameters given different test lengths, sample sizes, prior specifications, and/or correlations for these models, the key motivation is to provide researchers and practitioners with general guidelines when it comes to estimating a UIRT model and a multi-unidimensional IRT model. The results from the present study suggest that NUTS is equally effective as Gibbs sampling at parameter estimation under most conditions for the 2PL IRT models. Findings also shed light on the use of the two MCMC algorithms with more complex IRT models.

Apresentamos o tópico e a teoria de Modelos de Mistura de Distribuições, revendo aspectos teóricos e interpretações de tais misturas. Desenvolvemos a teoria dos modelos nos contextos de máxima verossimilhança e de inferência bayesiana. Abordamos métodos de agrupamento já existentes em ambos os contextos, com ênfase em dois métodos, o algoritmo EM estocástico no contexto de máxima verossimilhança e o Modelo de Mistura com Processos de Dirichlet no contexto bayesiano. Propomos um novo método, uma modificação do algoritmo EM Estocástico, que pode ser utilizado para estimar os parâmetros de uma mistura de componentes enquanto permite soluções com número distinto de grupos. / We present the topics and theory of Mixture Models in a context of maximum likelihood and Bayesian inferece. We approach clustering methods in both contexts, with emphasis on the stochastic EM algorithm and the Dirichlet Process Mixture Model. We propose a new method, a modified stochastic EM algorithm, which can be used to estimate the parameters of a mixture model and the number of components.

Apresentamos o tópico e a teoria de Modelos de Mistura de Distribuições, revendo aspectos teóricos e interpretações de tais misturas. Desenvolvemos a teoria dos modelos nos contextos de máxima verossimilhança e de inferência bayesiana. Abordamos métodos de agrupamento já existentes em ambos os contextos, com ênfase em dois métodos, o algoritmo EM estocástico no contexto de máxima verossimilhança e o Modelo de Mistura com Processos de Dirichlet no contexto bayesiano. Propomos um novo método, uma modificação do algoritmo EM Estocástico, que pode ser utilizado para estimar os parâmetros de uma mistura de componentes enquanto permite soluções com número distinto de grupos. / We present the topics and theory of Mixture Models in a context of maximum likelihood and Bayesian inferece. We approach clustering methods in both contexts, with emphasis on the stochastic EM algorithm and the Dirichlet Process Mixture Model. We propose a new method, a modified stochastic EM algorithm, which can be used to estimate the parameters of a mixture model and the number of components.

This thesis suggests a Bayesian vector autoregressive (VAR) model which allows for explicit parametrization of the unconditional mean for data measured at different frequencies, without the need to aggregate data to the lowest common frequency. Using a normal prior for the steady-state and a normal-inverse Wishart prior for the dynamics and error covariance, a Gibbs sampler is proposed to sample the posterior distribution. A forecast study is performed using monthly and quarterly data for the US macroeconomy between 1964 and 2008. The proposed model is compared to a steady-state Bayesian VAR model estimated on data aggregated to quarterly frequency and a quarterly least squares VAR with standard parametrization. Forecasts are evaluated using root mean squared errors and the log-determinant of the forecast error covariance matrix. The results indicate that the inclusion of monthly data improves the accuracy of quarterly forecasts of monthly variables for horizons up to a year. For quarterly variables the one and two quarter forecasts are improved when using monthly data.

Perfect Gibbs sampling is a method to turn Markov Chain Monte Carlo (MCMC) samplers into exact generators for independent random vectors. We show that a perfect Gibbs sampling algorithm suggested in the literature is not always generating from the correct distribution. (author's abstract) / Series: Research Report Series / Department of Statistics and Mathematics

A central problem in the bioinformatics is to find the binding sites for regulatory motifs. This is a challenging problem that leads us to a platform to apply a variety of data mining methods. In the efforts described here, a combined motif discovery method that uses mutual information and Gibbs sampling was developed. A new scoring schema was introduced with mutual information and joint information content involved. Simulated tempering was embedded into classic Gibbs sampling to avoid local optima. This method was applied to the 18 pieces DNA sequences containing CRP binding sites validated by Stormo and the results were compared with Bioprospector. Based on the results, the new scoring schema can get over the defect that the basic model PWM only contains single positioin information. Simulated tempering proved to be an adaptive adjustment of the search strategy and showed a much increased resistance to local optima.

The aim of this study was to explore the ranking model TrueSkill™ developed by Microsoft, applying it on various sports and constructing extensions to the model. Two different inference methods for TrueSkill was constructed using Gibbs sampling and message passing. Additionally, the sequential method using Gibbs sampling was successfully extended into a batch method, in order to eliminate game order dependency and creating a fairer, although computationally heavier, ranking system. All methods were further implemented with extensions for taking home team advantage, score difference and finally a combination of the two into consideration. The methods were applied on football (Premier League), ice hockey (NHL), and tennis (ATP Tour) and evaluated on the accuracy of their predictions before each game. On football, the extensions improved the prediction accuracy from 55.79% to 58.95% for the sequential methods, while the vanilla Gibbs batch method reached the accuracy of 57.37%. Altogether, the extensions improved the performance of the vanilla methods when applied on all data sets. The home team advantage performed better than the score difference on both football and ice hockey, while the combination of the two reached the highest accuracy. The Gibbs batch method had the highest prediction accuracy on the vanilla model for all sports. The results of this study imply that TrueSkill could be considered a useful ranking model for other sports as well, especially if tuned and implemented with extensions suitable for the particular sport.

In this paper we carry out fully Bayesian analysis of the general heterogeneity model, which is a mixture of random effects model, and its special cases, the random coefficient model and the latent class model. Our application comes from Conjoint analysis and we are especially interested in what is gained by the general heterogeneity model in comparison to the other two when modeling consumers' heterogeneous preferences. (author's abstract) / Series: Report Series SFB "Adaptive Information Systems and Modelling in Economics and Management Science"